Pain-related attention biases have a crucial role in the development and maintenance of chronic pain. Previous meta-analyses have demonstrated that individuals with chronic pain exhibit a sustained attentional biases toward pain-related stimuli. Several studies have also highlighted associations between the maintenance of pain-related attention biases and poorer long-term chronic pain outcomes. However, traditional measures used in previous studies including total fixation or duration indexes, cannot capture the dynamic nature of attention or variability in attentional processes between individuals. Some researchers have suggested that the attentional biases associated with chronic pain may exist at different stages of attention processing. Therefore, in order to gain a deeper understanding of the dynamic nature of visual attention biases toward pain-related stimuli and their potential predictive effects on responses to chronic pain, this study employed a time window segmentation analysis of eye movement data. Additionally, real pain stimuli were utilized in the visual task to elicit more authentic responses.
GPower3.1 was utilized to estimate the required sample size for this study; 49 participants were needed to detect an effect size (f) of 0.17 with a significance level (α) of 0.05 and a power of 95%. A total of 94 participants (69 women) experiencing chronic musculoskeletal pain (e.g., neck pain, shoulder pain, or low back pain), were recruited for this study. During the experiment, participants completed two tasks while their eye movements were recorded using an Eyelink 1000 eye tracker. The eye tracker had a sampling rate of 500 Hz, a spatial accuracy greater than 0.5°, and a resolution of 0.01° in the pupil-tracking mode. After receiving instructions, participants began the first task comprising 16 pairs of pain-neutral pictures and 16 pairs of neutral-neutral pictures, each measuring 11 cm × 10 cm. The viewing angle of each picture was 8.99° × 8.17°. In this task, picture pairs were displayed for 2000 ms, during which participants were instructed to freely view the pictures. Following the disappearance of the stimuli, a detection point appeared at the location of one of the pictures, and participants had to quickly and accurately judge the location of the detection point. Task 2 was identical to Task 1, exception that, no detection point was presented following the offset of picture pairs; instead, there was a possibility that an actual somatosensory pain stimulus would be delivered. Specifically, participants had a 25% chance of receiving a painful stimulus after each pain-neutral picture pair appeared while there was no chance a painful stimulus delivery after neutral-neutral picture pairs appeared. Participants were instructed to quickly and accurately determine whether or not they experienced a painful stimulus. At the start of the experiment, baseline data was collected, including the participants' chronic pain grade, pain catastrophizing scale scores, center for epidemiologic studies depression scores, and demographic information. Additionally, after a period of 6 months, the experimenters followed up with the participants to gather information on their chronic pain intensity and interference.
For Task 1 and Task 2, a 2 picture type (pain vs. neutral) ×4 epochs (0~500 ms, 500~1000 ms, 1000~1500 ms, and 1500~2000 ms) repeated measures ANOVAs assessing the attentional biases toward pain cues for patients with chronic pain was included. Then, bivariate correlation analyses evaluated the correlation between responses on baseline measures and follow-up levels of pain intensity and interference. Subsequently, significant attentional biases measures correlates of follow-up outcomes were assessed within separate machine learning and hierarchical standard multiple regression models for follow-up pain intensity and interference.
In Task 1, the main effect for picture type was found, F(1, 93) = 88.36, p< 0.001, η2p = 0.49. Participants displayed significantly longer attentional biases toward pain pictures (M = 49.63 ms, SE = 4.97) than neutral pictures (M = 2.10 ms, SE = 1.79). The main effect for epochs was found, F(3, 91) = 54.88, p < 0.001, η2p = 0.64. Participants displayed significantly longer attentional biases on the second epoch (M = 56.87 ms, SE = 4.00) than the first epoch (M = 10.15 ms, SE = 1.40), the third epoch (M = 22.00 ms, SE = 4.89) and the fourth epoch (M = 14.43 ms, SE = 4.87). The picture type × epochs interaction was found, F(3, 91) = 59.62, p < 0.001, η2p = 0.66. Patients with chronic pain displayed attentional biases toward pain pictures than neutral pictures during the first three epochs (0~500 ms, 500~1000 ms, and 1000~1500 ms) (ps < 0.001, see Figure 1A), but not during the fourth epoch (p = 0.39). In Task 2, the main effect for picture type was found, F(1, 93) = 83.76, p < 0.001, η2p = 0.47. Participants displayed significantly longer attentional biases toward pain pictures (M = 52.40 ms, SE = 5.30) than neutral pictures (M = 4.28, SE = 1.13). The main effect for epochs was found, F(3, 91) = 22.53, p < 0.001, η2p = 0.43. Participants displayed significantly longer attentional biases on the second epoch (M = 55.76 ms, SE = 6.67) than the first epoch (M = 6.64 ms, SE = 1.25), the third epoch (M = 32.11 ms, SE = 4.69) and the fourth epoch (M = 18.86 ms, SE = 4.40). The picture type × epochs interaction was found, F(3, 91) = 19.37, p < 0.001, η2p = 0.39. Patients with chronic pain displayed attentional biases toward pain pictures than neutral pictures during all the four epochs (0~500 ms, 500~1000 ms, 1000~1500 ms and 1500~2000 ms) (ps < 0.046, see Figure 1B).
By examining the magnitude of attentional biases across the four time windows in the two tasks, it was evident that attentional biases toward pain-related stimuli in patients with chronic pain were imbalanced. Attention was engaged in the first epoch of stimulus presentation (0~500 ms), reached its peak during the second epoch (500~1000 ms), and then gradually decreased during the third and fourth epochs (1000~1500 ms and 1500~2000 ms).
The bivariate correlation analyses revealed significant correlations between attentional biases toward pain pictures in the third and fourth epochs of both Task 1 and Task 2, and the follow-up levels of pain intensity and interference (rs > 0.22, ps < 0.04). Subsequent hierarchical standard multiple regression analyses revealed that attentional biases towards pain-related stimuli during the third and fourth epochs (1000~1500 ms and 1500~2000 ms) of both tasks independently predicted the persistence of chronic pain intensity and interference levels at a six-month follow-up, β = 0.01, t = 2.45, p = 0.02, ΔR2 = 0.05 (see Table 1, model a), β = 0.02, t = 3.39, p = 0.001, ΔR2 = 0.08 (see Table 1, model b), β = 0.01, t = 2.94, p = 0.004, ΔR2 = 0.07 (see Table 1, model c), β = 0.01, t = 2.65, p = 0.01, ΔR2 = 0.05 (see Table 1, model d), β = 0.01, t = 1.88, p = 0.06, ΔR2 = 0.03 (see Table 1, model e), β = 0.01, t = 1.91, p = 0.06, ΔR2 = 0.03 (see Table 1, model f), β = 0.01, t= 2.46, p= 0.02, ΔR2 = 0.04 (see Table 1, model g), β = 0.01, t = 1.88, p = 0.06, ΔR2 = 0.03 (see Table 1, model h). To investigate the stability of the above results, we utilized machine learning regression models. The machine learning regression model (Random Forest) found the consistent results (see Table 2). The correlations between the attentional biases towards pain pictures in the third and fourth epoch of Task 1 and Task 2, and the predicted chronic pain intensity and interference 6 months later generated by the machine learning regression model, was found to be significant, rs ≥ 0.23, ps ≤ 0.02. The fitting index between the predicted results of the machine learning regression model and the actual results is good, R2≥ 0.88, MSE ≤ 2.25.
In conclusion, attentional biases toward pain-related stimuli during the later stages (1000~1500 ms and 1500~2000 ms) predicted the maintenance of chronic pain intensity and interference levels at a six month follow-up. These effects were maintained even after controlling for baseline levels of pain intensity and interference and other baseline correlates of follow-up outcomes. The present study represents the first attempt to examine the impact of attentional bias towards pain-related stimuli on the maintenance of dysfunctional chronic pain outcomes from a dynamic perspective. These findings offer an explanation and valuable insights into attentional training, which holds significant importance in enhancing chronic pain management. Moving forward, training individuals to redirect their attention away from pain and associated cues during the later stages of attention may prove to be an effective approach for alleviating suffering due to chronic pain.
