Abstract
Objective To define biopsychosocial mechanisms of pain that go above and beyond disease activity and organ damage in systemic lupus erythematosus (SLE).
Methods We conducted a cross-sectional analysis of patient-reported data in a population-based registry of 766 people with SLE. Predictors of pain intensity and interference were examined using hierarchical linear regression. We built 2 main hierarchical regression models with pain intensity and interference as outcomes, both regressed on disease activity and organ damage. For each model, we sought to establish the relationship between pain outcomes and the primary exposures using sequential steps comprising the inclusion of each construct in 6 stages: demographic, socioeconomic, physical, psychological, behavioral, and social factors. We also conducted sensitivity analyses eliminating all overt aspects of pain in the disease activity measure and reestimated the models.
Results Disease activity and organ damage explained 32–33% of the variance in pain intensity and interference. Sociodemographic factors accounted for an additional 4–9% of variance in pain outcomes, whereas psychosocial/behavioral factors accounted for the final 4% of variance. In the sensitivity analyses, we found that disease activity and organ damage explained 25% of the variance in pain outcomes.
Conclusion Disease activity only explained 33% of the variance in pain outcomes. However, there was an attenuation in these associations after accounting for psychosocial/behavioral factors, highlighting their roles in modifying the relationship between disease activity and pain. These findings suggest that multilevel interventions may be needed to tackle the negative effect of pain in SLE.
Roughly 50–100 million Americans are living with ongoing pain, resulting in healthcare costs of $635 billion annually.1 An estimated 20 million people live with high-impact chronic pain with substantially restricted work, social, and self-care activities.1 Pain is the most frequently reported symptom in rheumatology, with the etiology ascribed to inflammation, joint degeneration, and central sensitization.2 Patients with systemic lupus erythematosus (SLE) rank pain as the most distressing symptom, above other symptoms such as fatigue, depression, sleep disturbance, weight gain, rashes, and forgetfulness.3 A previous survey revealed that 32% of patients with SLE listed joint and muscle pain and/or swelling as the symptom associated with the most negative impact on their lives.3 Despite treatment advances, pain remains the most prominent, underaddressed patient complaint. In a study of persistently frequent (≥ 3 visits per year) emergency department visits among patients with SLE, pain was coded as the chief concern for 50% of these visits.4
Chronic pain has lasting personal costs to patients, including poor quality of life, disability, social isolation, stress, and other psychosocial problems. The prevalence of chronic pain (defined as persistent pain that occurs on at least half the days for ≥ 6 months) and increased pain intensity (defined as the magnitude of experienced pain) vary by age, race/ethnicity, sex, educational attainment, and income.1,5,6 Pain interference (defined as pain that hinders major life activities) is also patterned by sociodemographic determinants.1,5 Lifestyle-related factors such as smoking and obesity, as well as comorbidities, are implicated in more severe pain manifestations.1,5 Despite the substantial effect of pain and extensive explorations of the determinants of health in SLE, the mechanisms of pain intensity and interference in SLE are not completely understood. We present a biopsychosocial approach (Supplementary Figure 1, available with the online version of this article) to explain pain intensity and interference as multidimensional, dynamic integration among disease-related, demographic, socioeconomic, physical, psychological, behavioral, and social constructs that reciprocally influence one another.7,8 We hypothesized that disease activity and organ damage would be associated with increasing pain intensity and interference in a cross-sectional sample of predominantly Black patients with SLE, along with other determinants. Quantifying the potential effect of modifiable behavioral, psychosocial, and SLE-related factors is important for the development of appropriate interventions to address the problem of pain in SLE.
METHODS
Study setting. The Georgians Organized Against Lupus (GOAL) cohort is a population-based cohort of individuals with a validated diagnosis of SLE supported by the Centers for Disease Control and Prevention. The overall aim is to examine the effect of sociodemographic and healthcare factors on outcomes that are relevant to patients, healthcare providers, and policy makers. Recruitment and data collection methods have been previously described.9 Consecutive annual sets of surveys have been administered to the GOAL cohort participants since 2012. All participants completed a self-report questionnaire to return by mail or to be completed online or by phone.
The Emory University Institutional Review Board, Grady Health System Research Oversight Committee, and Georgia Department of Public Health Institutional Review Board approved the GOAL study protocol. All participants provided informed consent.
Main exposures. Patient-reported responses from surveys through October 2015 to August 2017 were analyzed. The primary exposures of interest were disease activity and organ damage. Disease activity was measured using the Systemic Lupus Activity Questionnaire (SLAQ), a validated tool designed to be used in population-based studies outside the clinical setting, when physician assessment is not feasible.10,11 The SLAQ includes 24 questions to assess disease activity symptoms and signs (e.g., fatigue, fever, skin rashes, arthritis) in the past 3 months. Items are endorsed as “no problem,” “mild,” “moderate,” or “severe” and scored from 0 to 3. SLAQ scoring ranges from 0 to 44, with higher scores indicating greater SLE-related disease activity.
Organ damage accrual was measured using a validated self-administered version of the Brief Index of Lupus Damage.12 This tool measures cumulative organ damage in 12 organ systems that has been present for at least 6 months since SLE onset. Items are coded as present or absent, with scores ranging from 0 to 30 and higher scores indicating greater organ damage. This has been used in epidemiological studies and has shown to predict or correlate with important outcomes such as death, work loss, and depression.13,14,15
Main outcomes. The primary outcomes of interest were pain interference and pain intensity as reported at baseline and measured by the Patient Reported Outcomes Measurement System (PROMIS) adult short forms (SF).16
Pain intensity was measured using the following question from the PROMIS Global Health SF v1.0 (PROMIS Global-10): “In the past 7 days, how would you rate your pain on average?” An 11-item ordinal scale from 0 to 10 is provided to answer the question, with a higher score indicating more pain intensity.
Pain interference was measured using the PROMIS-SF Pain Interference 4a. This 4-item questionnaire uses a 5-item Likert scale rated from 1 (not at all) to 5 (very much) to quantify the effect of pain on daily activities, working around the house, participation in social activities, and completing household chores in the past 7 days. Raw scores were calculated by the PROMIS HealthMeasures Scoring Service and converted to t-scores. A t-score of 50 represents the reference population (mean 50, SD 10). A higher PROMIS t-score represents a greater presence of the concept being measured.
Covariates. We broadly defined 6 main constructs, as shown in Supplementary Figure 1 (available with the online version of this article).
Demographic factors included age at baseline, sex (female vs male), ethnicity (Black vs non-Black), and marital status (single vs married/living with a partner).
Socioeconomic factors included annual income in $10,000 increments, and educational attainment (high school or less, some college, and college and above).
Physical factors included quality of sleep, which was assessed with the PROMIS-SF Sleep Disturbance 8a. This is an 8-item measure of self-reported perceptions of sleep quality, depth, and restoration within the past 7 days. Patient-reported data were collected to measure BMI in kg/m2. Physical health was measured using a 5-point Likert scale question from the PROMIS Global-10: “In the past 7 days, how would you rate your physical health?” Answers were scored from 1 = poor to 5 = excellent.
Psychological factors included anxiety, depression, and anger. These domains were measured using PROMIS-SFs (Depression 8a, Anxiety 4a, and Anger 5a). These 3 measures have demonstrated clinical validity across a range of chronic health conditions.17
Behavioral factors included smoking, which was categorized as current vs not current. In addition, we used 3 scales from the Brief Coping Orientation to Problems Experienced (COPE) tool to measure negative mechanisms of coping (substance/alcohol use, self-blame, and denial) along with 1 scale to measure coping with religion. The tool has good psychometric properties to measure coping strategies used in everyday life or in distressing situations.18,19
Social factors included emotional support and social isolation, which were measured with the following PROMIS-SFs: Social Isolation 6a assessed perceptions of being avoided, excluded, detached, or disconnected from, or unknown by others; and Emotional Support 4a measured feelings of being cared for and valued as a person and having confidant relationships.20 In addition, a modified version of the Everyday Discrimination Scale was used to measure various forms of interpersonal mistreatment in participants’ day-to-day lives over the previous 12 months. Examples include being “treated with less respect than other people” and “treated as if you are not smart.” The 10 items on the scale are framed in the context of general mistreatment, without reference to race/ethnicity and other demographics. Responses were assessed with a 4-point scale (1 = never, 2 = rarely, 3 = sometimes, 4 = often), which was summed and averaged for a final score. The everyday discrimination scale has been widely used across samples of Black, White, and Chinese participants,21,22,23,24 and has shown high levels of internal consistency as well as convergent and divergent validity.21,25 We also measured unmet financial needs using the analogous 4-item scale included in the Conger Financial Strain measure.26 The scale assesses specific needs that cannot be met due to financial hardship (e.g., not enough money to buy the [home/clothing/food/medical resources] we need).
Statistical analysis. The data were analyzed using SAS version 9.4 (SAS Institute). Baseline characteristics were obtained using summary statistics. Continuous variables were summarized using means, SDs, and medians. The unadjusted associations between covariates and the exposures and outcomes were estimated using linear regression. We used G*Power to conduct a posthoc calculation of required sample size for a linear multiple regression to test the increase in R2 with a power of 95% and a model with 23 predictors, with small-to-medium effect size (0.05).27 The required sample size was 664. We used all of the available data of 766 patients, making this study more than adequately powered.
Predictors of pain intensity and interference were examined using hierarchical linear regression. We built 2 main regression models: both pain intensity and pain interference regressed on disease activity and organ damage. Each regression model sought to establish the relationship between pain outcomes and the primary exposures using hierarchical steps comprising the inclusion of each construct in 6 stages: demographic (sex, age, marital status, and race/ethnicity), socioeconomic (annual household income and educational attainment), physical (sleep disturbance, BMI, and global health–physical), psychological (anxiety, depression, and anger), behavioral (smoking, coping with religion, coping with substance/alcohol abuse, and coping with denial), and social (emotional support, social isolation, financial strain, and discrimination). In advance of the analyses, we specified an a priori framework (Supplementary Figure 1, available with the online version of this article) and a logically determined priority of each construct based on clinical expertise on how pain would be evaluated in a clinical scenario. Thus, entered into the first stage were biological or disease-related constructs: disease activity and organ damage. At Stage 2, demographic characteristics were entered, followed by socioeconomic factors at Stage 3, physical factors at Stage 4, psychological factors at Stage 5, behavioral factors at Stage 6, and social factors at Stage 7. Our goal was to determine whether newly added constructs showed significant improvement in the proportion of explained variance in pain intensity and interference by the models (R2) over disease-related constructs.
Sensitivity analyses. To account for possible collinearity between disease activity and pain outcomes, we examined the effect of a modified SLAQ score without 6 pain-related measures: stomach pain, chest pain, muscle pain, headache, joint swelling, and joint pain. We modeled the 2 pain outcomes on the modified SLAQ score with the other variables to determine if there were changes in the inferences from the original SLAQ score. We also examined the individual contributions of constitutional, mucocutaneous, organ system, and musculoskeletal symptoms in SLAQ on pain outcomes. We replaced the full SLAQ with the scores for each of the individual contributions in Models 1 and 7.
RESULTS
Baseline characteristics. There were 766 participants, of which 93% were female (Table 1). The mean age was 48 years. A majority were Black (82%), nonsmokers (87%), single (62%), and had a reported annual income of < $40,000 (67%). The mean disease activity score was 16.2 and the mean organ damage score was 2.5. The means and SDs were 5.3 ± 2.9 for pain intensity and 54.8 ± 10.2 for pain interference. Means and SDs of PROMIS measures were 55.7 ± 9.9 for sleep disturbance, 51.3 ± 9.3 for emotional support, 51.8 ± 10.4 for depression, 5.8 ± 10.2 for anxiety, 49.0 ± 10.9 for social isolation, 2.5 ± 0.9 for physical health, and 52.8 ± 12.3 for anger. Other means and SDs were 1.8 ± 0.6 for everyday discrimination and 9.2 ± 3.3 for financial strain.
Baseline characteristics of patients in the Georgians Organized Against Lupus cohort.
Bivariate analyses. The unadjusted associations between predictors, main exposures, and outcomes are shown in Table 2. The following characteristics were significantly associated with higher pain intensity: single marital status, Black ethnicity, lower household income, and lower educational attainment. Age and sex did not have a significant association with pain intensity. Higher levels of sleep disturbance and BMI were associated with higher pain intensity. Better self-reported physical health was associated with lower pain intensity. Participants who reported higher levels of anxiety, depression, and anger were also more likely to report higher pain intensity. Smokers were more likely to report higher pain intensity in comparison to nonsmokers. Higher reported levels of coping with religion or spirituality were associated with increased pain intensity. Increased negative coping characteristics (substance/alcohol use, self-blame, and denial) were associated with higher pain intensity. However, the association between coping and substance/alcohol use was not significant. Higher levels of emotional support were associated with decreased pain intensity; conversely, higher levels of social isolation, financial strain, and discrimination were associated with increased pain intensity. These bivariate associations were similar for pain interference and disease activity (except for age). The following were significantly associated with organ damage: age, household income, sleep disturbance, physical health, anxiety, depression, as well as coping with self-blame, social isolation, and financial strain.
Factors associated with pain outcomes and disease-related measures in individuals with SLE in the unadjusted linear regression.
Hierarchical regression modeling. The association between pain intensity and disease activity and organ damage adjusted for all the constructs in the full model explained up to 53% of the variance in pain intensity (Table 3, Table 4, Figure 1). Disease activity and organ damage explained 31% of the variance in pain intensity, with increased disease activity correlating with increased pain intensity. However, the association between organ damage and pain intensity was not significant in the 7 models. The magnitude of the association between disease activity and pain intensity was attenuated going from the unadjusted model (β 0.179, 95% CI 0.159–0.199) to the full adjusted model (β 0.106, 95% CI 0.081–0.130). Demographic and socioeconomic factors accounted for an additional 9% of variance in pain intensity, with older age, male sex, Black ethnicity, lower income, and lower educational attainment showing significant association with increased pain intensity in the fully adjusted model. Sleep disturbance, BMI, and physical health explained an additional 9% of the variance in pain intensity; all 3 measures remained significant in the fully adjusted model. Increased sleep disturbance and increased BMI were associated with increased pain intensity, while better self-reported physical health was associated with lower pain intensity. Psychological, behavioral, and social factors accounted for the final 4% of the variance in pain intensity. Only anger and coping with denial remained significant in the fully adjusted model.
Summary of the hierarchical linear regression models.
Hierarchical linear regression analyses of 7 constructs with pain intensity as outcome.
A representation of the proportion of variance in pain intensity and pain interference explained by 7 constructs in the biopsychosocial framework. * Corresponds to results of the sensitivity analysis that tested the model using the modified SLAQ. The modified SLAQ score removes stomach pain, chest pain, headache, joint swelling, joint pain, and muscle pain. SLAQ: Systemic Lupus Activity Questionnaire.
The findings of the hierarchical regression analyses of pain interference on disease activity and organ damage adjusted for the other constructs are shown in Table 5. Approximately 54% of the variance in pain interference was explained by the constructs. Disease activity and organ damage explained 33% of the variance in pain interference; however, only disease activity remained significantly predictive of pain interference in the fully adjusted model (Table 3, Figure 1). The effect size of the association between disease activity and pain interference was attenuated going from the unadjusted (β 0.593, 95% CI 0.528–0.659) to the fully adjusted model (β 0.261, 95% CI 0.180–0.341). Demographic and socioeconomic factors explained an additional 4% of the variance in pain interference. Only older age and lower educational attainment remained significant in the final model. BMI, sleep disturbance, and physical health explained a further 12% of the variance in pain interference, with all 3 measures remaining significant in the final model. Psychological, behavioral, and social factors accounted for the final 4% of the variance in pain interference; however, only anger and coping with self-blame remained significant in the fully adjusted model.
Hierarchical linear regression analyses of 7 constructs with pain interference as outcome.
Sensitivity analyses. Supplementary Tables 1 and 2 (available with the online version of this article) show the findings of the sensitivity analyses using the modified disease activity measure in which all of the pain-related items in the disease activity score were removed. While the inferences from Tables 4 and 5 hold, the modified disease activity measure combined with organ damage explained 25% of the variance in pain intensity and interference, which is less than the 32–33% of the variance in the original models using the full disease activity scores. After combining the other constructs, the full models explained 51–52% of the variance in both outcomes. Table 3 and Figure 1 show that the other constructs (physical, psychological, behavioral, and social) explain more of the variance in pain outcomes after accounting for the pain items in disease activity.
In addition, individual contributions of SLAQ symptoms to the variance in pain outcomes varied (Supplementary Tables 3 and 4, available with the online version of this article). Consistent with Supplementary Tables 1 and 2, we found that the models with constitutional, mucocutaneous, and organ system symptoms explained 17–27% of the variance in pain outcomes, while the models with musculoskeletal symptoms contributed 38–39% of the variance in pain outcomes.
DISCUSSION
Our findings highlight the complex and dynamic interactions of 7 constructs in the perceptions of pain among patients with SLE. We found that individuals reporting increased disease activity also reported higher pain intensity and interference. There was an attenuation in these associations after accounting for other constructs, highlighting their roles in modifying the relationship between disease activity and pain intensity. The biopsychosocial model has been used in understanding the correlates of pain and fatigue in rheumatoid arthritis, multiple sclerosis, osteoarthritis, and sickle cell disease.28,29,30,31,32 These studies found significant roles of disease activity/inflammation and psychosocial factors, similar to our findings. Our findings are consistent with a cohort of patients with SLE in which disease activity measured by the Systemic Lupus Activity Measure (SLAM) and Systemic Lupus Erythematosus Disease Activity Index was more than twice as high in a group of patients reporting high levels of pain compared with those reporting low levels of pain.33
Our understanding of the mechanisms of inflammation that affect pain in SLE (and in general) is evolving. However, we confidently used SLAQ as a surrogate for inflammation because higher disease activity is correlated with higher levels of serum biomarkers of inflammation.34 When we omitted pain-related items from the SLAQ score, the association between disease activity and pain intensity persisted. Further, we examined the individual contributions of constitutional, mucocutaneous, organ system, and musculoskeletal symptoms in SLAQ and found that mucocutaneous and constitutional symptoms contributed to pain to a lesser extent than musculoskeletal symptoms. Greco, et al35 showed similar findings in their study investigating chronic pain clustering in SLE. The pain clusters did not change significantly when pain symptoms were omitted from the revised SLAM.35 All these findings suggest that individual, behavioral, and social factors can potentially exacerbate or attenuate the intensity of pain caused by inflammatory mediators in patients with SLE with active disease.
We found that demographic and lifestyle constructs play crucial roles in pain severity. Our finding that older age was independently associated with increased pain intensity and interference is consistent with other epidemiological studies that have found an age-related increase in the prevalence of chronic pain.1 We found that Black patients reported higher pain intensity than non-Black patients. These findings run parallel with sex and racial/ethnic differences in the development and outcomes of SLE.36,37 Males also seem to have more severe disease compared with females.38 However, these findings are contrary to what is known in the general population, in which females and White individuals have higher prevalence of chronic pain, suggesting that pain in SLE has a different epidemiological profile.1 We found that higher socioeconomic status was protective for pain outcomes. In patients with SLE, socioeconomic status is associated with lower disease activity and damage accrual.39 In the general population, socioeconomic disadvantage is associated with almost every aspect of poorer health, including increased morbidity and decreased life expectancy. Not surprisingly, it is also consistently associated with increased risk for pain.40
A growing body of research suggests that obesity and poor sleep quality, which have been associated with pain outcomes, may play important roles in SLE outcomes. Several studies have linked obesity with disease activity, increased risk of renal disease, cardiovascular complications, fibromyalgia, depression, poorer functional capacity, and decreased quality of life in patients with SLE.41,42,43,44 Obesity is hypothesized to cause pain through excess mechanical stresses. Due to its proinflammatory state, obesity is also a marker of increased functional and psychological complications in chronic pain.45 Moreover, recent reports along with our findings suggest that complex pathways may link sleep disturbance and pain along with depression and cognitive symptoms.46 Thus, further research is needed to determine whether interventions specifically developed to tackle those factors may effectively improve pain outcomes in SLE populations.
In addition, we found that psychological, behavioral, and social factors were associated with pain intensity and interference in the unadjusted models. In the fully adjusted models, these associations were no longer significant. When we omitted overt manifestations of pain from the disease activity measure (as seen in the sensitivity analyses), we found that psychosocial and demographic factors accounted for more variance in pain intensity and interference, highlighting their importance in pain outcomes. The psychosocial burden of SLE is immense. Compared to the general population, patients with SLE report impaired quality of life and more fatigue, fibromyalgia, anxiety, and depression.47 These psychosocial factors also disproportionately increased with disease activity and organ damage. Future studies elucidating the role of psychosocial factors in pain are needed for the appropriate design of targeted interventions for patients with SLE.
Our study has some limitations. First, we cannot rule out residual and unmeasured confounding due to variables not capturing the complete essence of constructs such as socioeconomic and psychological factors. Second, the survey did not capture information on fibromyalgia. Third, the cohort may not be generalizable to the US population of patients with SLE. However, as SLE affects predominantly Black women—comprising 43% of SLE cases in the US according to prevalence estimates—our findings are generalizable to a large US population, particularly the southeastern region of the country. Fourth, given the different assessment periods for the patient-reported outcomes and the cross-sectional nature of the design, we were unable to investigate the temporal relation between pain and associated factors, which in turn may affect the interpretation of our results. Finally, our findings may be confounded by the ordering of the constructs in the regression models. The choice of the order of the constructs was made prior to model building, based on the assumption that the main predictors of pain in SLE are disease activity and organ damage. Our findings suggest that a range of multidimensional interventions to reduce pain should continue to be directed to disease-related factors, as clinically indicated. This is the single construct that explained the largest proportion of variation in pain. However, this proportion never exceeded 33% in our models, with significant remaining proportions explained by demographic, socioeconomic, physical, psychological, behavioral, and social factors. These are constructs that are not modified directly by immunologic interventions but speak to other mechanisms that should be considered when aiming to improve the lives of those with SLE. This may include programs focusing on reducing obesity and improving sleep hygiene, as well as psychological interventions addressing anxiety, depression, and anger. Behavioral interventions, such as smoking cessation and improving coping skills, should be considered. Social programs to improve emotional support and decrease isolation may be helpful. Validated and culturally appropriate self-management programs will be an important tool in this space.
Our study has several strengths. To our knowledge, this was the first study to present a biopsychosocial model for pain intensity and interference in SLE. We leveraged a large sample of patients to make these inferences. However, this was a cross-sectional study and we were unable to answer questions about causality. Therefore, future studies of longitudinal design are needed to ascertain causality. In addition, while our models elucidate a proportion of factors, there is also a significant proportion of variance that remains undefined. Further research is needed to confirm these observations as well as to identify other factors that continue to remain undefined in our models.
Footnotes
The authors did not receive any financial support or other benefits from commercial sources for this work, or any other financial interests which could create a potential conflict of interest or the appearance of a conflict of interest with regard to this work.
- Accepted for publication November 20, 2020.
- © 2021 The Journal of Rheumatology
REFERENCES
ONLINE SUPPLEMENT
Supplementary material accompanies the online version of this article.
