Biopsychosocial Factors Associated With Pain and Pain-Related Outcomes in Adults and Children With Sickle Cell Disease: A Multivariable Analysis of the GRNDaD Multicenter Registry

Participants and Procedures

This study used data from GRNDaD³⁰ research registry (https://grndad.org), a multicenter registry that aims to prospectively collect relevant data on individuals with SCD, including patient-reported outcomes (PROs), labs, imaging, and other clinical data. Participants of all ages are eligible for inclusion in the GRNDaD registry if they have documentation of laboratory diagnosis of SCD. For this current study, participants were included if they met the following criteria: 1) diagnosis of SCD with genotypes SS, $\text{Sβ}$-thalassemia, or SC, 2) age 8 years or older, and 3) completion of at least 1 pain or pain-related PRO relevant to the outcomes of interest at the time of data collection (April 2022). We restricted the age to 8 years or older because SCD pain is not a significant burden in early childhood.

Fifteen SCD centers had contributed data to GRNDaD at the time of this study. Regulatory approval for GRNDaD was obtained from a single institutional review board (Johns Hopkins University School of Medicine). GRNDaD study personnel at each SCD center consent and enroll eligible participants, and collect baseline data, which is stored in a secure manner via the Research Electronic Data Capture database.

GRNDaD centers are required to collect sociodemographic, laboratory, and clinical data that are relevant to SCD care and complications, but pain-related PRO collection varies across centers (and is not required as part of the minimum data set). The availability of research funding to collect and transfer data to Research Electronic Data Capture is the primary determinant of whether a center includes additional PROs beyond what is required for participation in the registry. At the time of this analysis, 2104 participants, 8 years old or older, were enrolled in the GRNDaD registry. Of these, 723 participants completed at least 1 pain-related survey of interest—the Adult Sickle Cell Quality of Life Measurement Information System (ASCQ-Me)³¹ pain subscales, the Brief Pain Inventory Short Form (BPI),³² or the Patient-Reported Outcomes Measurement Information System Physical Function Short Form 8b³³ (PROMIS physical function). A total of 657 participants from 10 sickle cell centers had completed the ASCQ-Me pain subscales, which assess pain impact (pain impact sample [Fig 1]). A total of 403 participants completed the BPI measure of pain intensity (the pain severity sample), and 267 participants completed the PROMIS physical function questionnaire (the physical function sample), respectively (Fig 1). Two hundred and three participants had completed all 3 questionnaires (ASCQ-Me, BPI, and PROMIS physical function) and, thus, were included in all 3 samples. The pain impact sample represents the primary sample for the analysis that is described in this study. Separate analyses were conducted for the pain severity sample and the physical function sample.

Measures

Self-report measures included sociodemographic information (ie, education and income) on enrollment to the registry and specific measures of pain severity, pain locations, frequency of sickle cell pain episodes, pain impact, emotional functioning, social functioning, and use of hydroxyurea. The presence of chronic pain was obtained via a self-report measure that was based on a published definition of chronic SCD pain.⁷ For participants younger than 18 years old, PRO data was obtained by parental proxy. Clinical and laboratory measures such as hemoglobin and fetal hemoglobin percentage (HbF%) were obtained from participant medical records by GRNDaD study personnel. Laboratory values and PRO measures were obtained within the same calendar year, and the majority of participants had lab values collected within 1 month of PRO collection. For participants who completed multiple pain-related PROs since enrollment in GRNDaD, we used the most recent PRO available in the database.

Pain Outcomes

We examined pain and pain-related outcomes that correlate with early mortality or poor quality of life in SCD,³ and/or are included in consensus guidelines on chronic pain studies in adults and children.^34,35 The primary outcome was pain impact, as measured by the ASCQ-Me³¹ Pain Impact Short Form, a 5-item subscale that assesses pain interference over a 7-day recall period.³¹ This subscale has been demonstrated to strongly correlate with the PROMIS Pain Interference and Pain Behavior questionnaire and, thus, is a disease-specific measure of pain interference.³⁶ Raw scores were converted to T-scores using conversion tables in the ASCQ-Me User’s Manual,³⁷ with scores ranging from 0 to 100. A higher T-score indicates “better” health or less pain impact.

Given the multidimensional nature of the SCD pain experience, we also explored 2 additional pain-related outcomes—numeric pain severity and physical function—using the BPI questionnaire and PROMIS Physical Function Short Form 8b. The BPI is a self-report questionnaire³² that has been administered across a variety of pain conditions. To assess pain severity in our sample, we used a composite of the 4 pain items of the severity subscale (ie, a mean severity score), as recommended by the BPI developers.³⁸ PROMIS Physical Function Short Form 8b is an 8-item questionnaire that measures physical impairments that impact a person’s ability to carry out daily activities. Total raw scores were calculated and converted to T-scores using the applicable score conversion table in the PROMIS Physical Function Scoring Manual.³⁹

Predictor Variables for Multivariable Modeling

Statistical Analyses

Demographics and baseline characteristics of participants were summarized. A multivariable linear regression model was created for our primary outcome of interest—pain impact—using the approach outlined in Frank Harrell’s textbook, Regression Modeling Strategies.⁴⁰ The predictor variables included in the models were age, SCD genotype, gender, income, education level, hemoglobin, HbF%, BPI pain variability, BPI number of pain locations, ASCQ-Me emotional functioning, ASCQ-Me social functioning, ASCQ-Me pain episodes, and use of hydroxyurea. Nonreported variables were assumed to be missing at random. Multiple imputation of each sample was conducted to address missing data, which averaged 10 to 20%, depending on the variable and participant sample. Some variables, such as HbF% had high missing rates (greater than 30%) in the pain impact sample but were ultimately included in the analysis due to sufficient accuracy of the predictors used in the imputation (imputation R² = .71). A summary of the imputation performance and missing data is provided in Supplementary Table 1. Multiple imputation was performed with aregImpute, which uses a chained equation approach with bootstrap samples, each of which is imputed using weighted predictive mean matching prior to modeling. As each sample was defined by the availability of the outcome measure, none of the outcomes were imputed, albeit the outcomes were used to impute missing predictors to reduce bias.^40,41 To avoid missing complex, nonlinear relationships between a predictor and an outcome, continuous variables were modeled using restricted cubic splines, with 5 knots for age, hemoglobin, and HbF%, and 3 knots for pain variability, emotional functioning, social functioning, and number of pain locations. Income and education level were simplified into fewer categories to allow for better imputation. Given the relatively few predictors, the large size of each sample, and the goal of effect estimation, all the prespecified predictors were included in the final model with no data reduction. Internal validation of the model was conducted to assess for the degree of overfitting. The relative importance of each predictor variable to the model was derived from the Wald Chi-square analysis of variance of the full model compared to the model without the potential predictor, subtracting the degrees of freedom used by the predictor. Partial effect graphs of the most important predictors in the model were presented, with the other predictors set at the median (continuous predictors) or mode (categorical predictor) values. Statistical analysis was conducted with the R statistical program (R Foundation for Statistical Computing, Vienna, Austria) version 4.2.2, using the RMS package for multiple imputation, linear regression modeling, partial effect prediction, and validation. Graphs were generated with the ggplot2 package in R. P values and 95% confidence intervals were presented graphically or reported for all statistical tests. A full list of the R packages, model specifications, and regression coefficients is available in the Supplementary document.

Multivariable regression models for pain severity and physical function samples were also generated using the statistical approach outlined above.

Description of Sociodemographic and Clinical Characteristics

The pain impact sample was 60% female with a median age of 34 (interquartile range [IQR] 26–42 years). Most of the participants had sickle cell anemia (HbSS genotype [n = 397, 67%])—the predominant genotype in SCD—and the remainder had sickle hemoglobin C disease (HbSC genotype [n = 123, 21%]) or sickle beta-thalassemia disease (${\text{HbSβ}}^{+}$ [n = 42, 7.1%] or ${\text{Hbβ}}^0$ [n = 30, 5.1%] genotypes). Fifty-eight percent of the participants were on hydroxyurea. The overwhelming majority reported an annual income of less than $41,320, with 44% reporting an annual income of less than $12,060. In terms of clinical pain characteristics, chronic pain was reported in 64% of the sample at the time of enrollment into the registry. The median number of pain locations reported was 6 out of 22 locations (27%). A large number (n = 240, 44%) of participants reported having 4 or more sickle cell pain episodes in the prior 12 months. Most participants had at least 1 hospital admission during the year of enrollment into GRNDaD, and approximately 22% had 4 or more hospital admissions in the enrollment year. The median pain impact score in the sample was 51 (IQR 44–64), and the median PROMIS physical function score was 43 (IQR 38–49). Table 1 details additional clinical and sociodemographic characteristics of the pain impact sample. Descriptive statistics for the pain severity and physical function samples were comparable to the pain impact sample and are presented in Table 2 and Supplementary Table 2.

Variables Associated With High Pain Impact

Overall, for pain impact, our multivariable regression model accounted for 58% of variance (${\text{R}}_{\text{adj}}^2=.58$). In our analysis of the relative contribution of each variable in the model for pain impact, we found that social functioning (Wald X² = 87.47; P < .001), emotional functioning (Wald X² = 57.86; P < .001), income (Wald X² = 20; P < .001), age (Wald X² = 15.60; P = .004), and number of sickle cell pain episodes (Wald X² = 13.98; P = .007) had the strongest association with pain impact (Fig 2). The relationships between these variables and pain impact are visually depicted in the prediction graphs found in Fig. 3A to 3E. Regression coefficients for all the variables examined in the model are in the Appendix.

In this multivariable analysis, we found that better social and emotional functioning was associated with lower pain impact (Fig. 3A and 3B). Changes in pain impact were most pronounced at social and emotional functioning T-scores of 45 to 60. Predicted changes in pain impact were minimal outside of this score range for social and emotional functioning.

In our model of pain impact, increasing age was associated with worse pain impact (lower T-scores) (Fig 3D). Examination of the predicted change in pain impact with increasing age showed that changes in pain impact were most pronounced during late adolescence and young adulthood (age 20 to early 30s). Pain impact increases steadily (decreasing T-scores) in late adolescence and young adulthood. From the early to mid-30s, pain impact decreases slightly (increasing T-scores) compared to young adulthood. After age 40, age-associated changes in pain impact are minimal.

Variables Associated With Increased Pain Severity and Worse Physical Function

For the pain severity sample, our multivariable regression model accounted for 48% of variance (${\text{R}}_{\text{adj}}^2=.48$). We found that pain variability (Wald X² = 38.76; P < .001), age (Wald X² = 17.39; P = .002), social functioning (Wald X² = 15.34; P = < .001), and number of pain locations (Wald X² = 11.58; P = .003) had the strongest association with pain severity in the model (Supplementary Fig 1). Thus, short-term pain variability, increasing age, greater number of pain locations, and lower social functioning were most associated with higher pain severity.

For the physical function sample, our multivariable linear regression model accounted for 55% of the variance in physical function (${\text{R}}_{\text{adj}}^2=.55$). Age (Wald X² = 32.33; P < .001) and social functioning (Wald X² = 18.25; P < .001) were the strongest contributors to physical function in this model (Supplementary Fig 2). Thus, increasing age and lower social functioning had the strongest association with lower physical function.

The prediction graphs for the significant variables in the pain severity and physical function models are presented respectively in Supplementary Fig. 3A to 3D and 4A and 4B. The Appendix contains the regression coefficients for all the variables examined in the pain severity and physical function models.

Pain Impact Sample Findings

We found that over 60% of our pain impact sample had chronic pain and that the median physical function score in the overall sample was below the standardized population score of 50. These findings are consistent with prior studies highlighting that chronic pain is a significant burden in SCD and is associated with poor physical functioning and disability.^42,43

Our multivariable model explained 58% of the variance in pain impact. The strength and findings of this model suggest that psychosocial variables (ie, social functioning and emotional functioning) are more strongly associated with pain impact in individuals with SCD than biological variables such as SCD genotype, hemoglobin, or HbF%. SCD genotype is considered the main determinant of disease severity. The HbSS genotype is generally believed to be the most severe SCD genotype.^4,44 Hemoglobin and fetal hemoglobin are also modifiers of disease severity in SCD. However, in our model, we found that SCD genotype, hemoglobin, and HbF% were less strongly associated with pain impact compared to social and emotional functioning, income, age, and frequency of pain episodes. Selection bias may account for some of these findings, as individuals with the HbSC genotype who are doing well may be less likely to be receiving care at an SCD center and, thus, be enrolled in GRNDaD. Nevertheless, the results of our multivariable regression modeling agree with the tremendous variability in pain experiences that has been observed clinically among individuals with the same SCD genotype.⁸ Taken together, these findings underscore the notion that, while SCD genotype, hemoglobin, and fetal hemoglobin are important markers of disease severity and disease-specific complications, other factors, such as patient psychosocial characteristics, may be stronger contributors to pain impact in SCD. We believe that this represents a novel finding that contributes to the literature on SCD pain. Of note, we were limited to the biological/laboratory variables routinely reported in GRNDaD. Indeed, other biological variables associated with increased pain intensity or frequency in SCD (eg, co-inherited $\text{α}$-thalassemia, reticulocyte count, mean corpuscular volume, and serum lactic dehydrogenase values)⁴⁴ were not included in our models due to the paucity of reliable data.

Our findings add to the growing body of literature that highlight the influence of psychosocial variables on pain outcomes in SCD. Prior studies have noted that poor pain self-efficacy, poor pain coping, and high pain catastrophizing are associated with increasing pain intensity and pain-related disability.^17,19,20 Other studies have also noted that elevated fatigue, anxiety, and depression are associated with high self-reported pain.^15,21 We believe that our findings extend beyond these prior studies by highlighting that psychosocial variables may have a comparable or even greater influence on pain impact than biological markers of disease severity, such as genotype or hemoglobin. Further exploration of the specific psychosocial variables that hold the most importance is warranted in future prospective studies. Nevertheless, our findings, combined with prior research, suggest that to attain comprehensive pain management in SCD, targeted pain interventions will need to expand to include psychosocial variables alongside biological variables.

Age is a known sociodemographic predictor of disease severity and pain in SCD. Prior reports on SCD have noted a relationship between increasing age and chronic pain intensity and chronic pain disability.^19,21,45 Additionally, health care utilization is highest in the 20 to 30 age group, and overwhelmingly, most of these health care encounters involve pain management.⁴⁶ Multiple factors contribute to the vulnerability of the adolescent and young adult (AYA) population, including the lack of structured transition from pediatric to adult care, increase in disease complications as individuals age, and transition from acute to chronic/persistent pain.^6,7 Our results provide greater insights into the pain burden in the AYA sickle cell population. In this study, we used restricted cubic splines for age, which allowed us to model the complex, nonlinear relationships between pain impact and age (as well as pain severity and age in a separate patient sample). For our primary model of pain impact, we found that the steepest age-related change in pain impact occurred during the 20 to early 30s age group, and pain impact was worse during this period. Consistent with prior studies that report high health care utilization in the AYA sickle cell population, these results emphasize that the AYA group is a critical and vulnerable subgroup of individuals living with SCD. The need to develop interventions that specifically target this subgroup is urgent.

Pain Severity Sample Findings

In addition to psychosocial variables, we examined 2 pain-specific variables (ie, number of pain locations and short-term pain variability), which, to our knowledge, have only been studied in 3 prior predictive SCD studies.^20,47,48 In our pain impact model, neither pain variability nor the number of pain locations appeared to be as important to pain impact as the psychosocial variables. However, in the model for pain severity, these 2 pain-specific variables had strong associations with pain severity. Given that our models for pain impact and pain severity included overlapping, but not identical, patient samples, we cannot make direct comparisons between the 2 models. Age had a significant complex, nonlinear relationship with pain severity, which was greatest in the 20-to-early 30s age group, once again highlighting that the AYA population is the SCD subgroup that is most vulnerable to pain.

Physical Function Sample Findings

Increasing age had the strongest association with worsening physical function compared to the other variables in the model. Prior research has shown that, in SCD, pain is the dominant symptom during young adulthood, while during older ages, complications secondary to organ dysfunction, such as chronic kidney disease and pulmonary hypertension, predominate.⁴ Thus, organ complication, rather than pain, is likely the driving force for functional impairment in older individuals (ie, greater than 40 years) with SCD.

Comment on Overall Pattern of Findings From Pain Impact, Pain Severity, and Physical Function Models

The prevalence of our pain-specific variables across the 3 patient samples provides some insight into potential mechanisms that contribute to chronic SCD pain. We found that pain in multiple locations was common across all 3 samples, with the number of pain locations averaging greater than 6 (or 27–30% of 22 locations). This is comparable to findings from the Pain in Sickle Cell Epidemiology Study, a longitudinal study of pain in adults with SCD, which showed that the average number of pain locations reported among the study cohort was 3.3/16 locations (25%).⁴⁷ Zempsky et al⁴⁸ also found that over 20% of adolescents hospitalized for a VOE reported pain in 7 or more unique body locations. Our findings, along with these from prior studies, suggest that widespread pain is common in SCD and provide a rationale for further investigation of features of centralized pain disorder in SCD. A high prevalence of central pain in SCD warrants additional studies on the role of centrally acting analgesics, non-pharmacologic behavioral interventions, and support for social needs in managing SCD-related pain.²⁹

This study leveraged data collected prospectively from a research registry to analyze pain outcomes in large samples of individuals living with SCD. Unlike other chronic pain disorders, SCD is a rare disease in the United States, and recruiting from small populations is challenging, particularly when considering the types of epidemiologic studies required to understand pain outcomes in SCD. The insight gained from our results highlights the important role that research registries, such as GRNDaD, can play in addressing gaps of knowledge in SCD pain. However, to fully leverage the potential of registries like GRNDaD, standardization and systematic collection of PROs within the clinical setting are required. Considering this study’s findings, we recommend that all participating centers collect PROs capturing data on participants’ mood and pain distribution. PROMIS measures, such as the depression and anxiety scales, would be appropriate given the presence of validated adult and pediatric versions, which would facilitate comparisons across the lifespan. Future efforts should also foster collaboration with SCD centers and streamline the integration of PRO data collection into routine clinical practice and electronic medical records. This concerted approach will ensure that chosen PROs hold both clinical significance and practical feasibility for implementation across different sickle cell centers. By addressing these improvements, registries, such as GRNDaD, can be leveraged to study pain trajectories and outcomes in a longitudinal manner and, thus, advance our understanding of pain outcomes in SCD.

Strengths and Limitations

The results of this study should be interpreted with consideration of a few limitations. The study was a cross-sectional analysis of registry data, and thus, we are unable to establish causality or definitively predict future pain outcomes. Further, the study is limited by the data available in the GRNDaD registry, and as it was also reliant on patient self-report or parental proxy for reporting of outcomes, it may be subject to recall and nonresponse bias. The PROs used in this study (BPI, ASCQ-Me, and PROMIS Physical Function) are adult measures, and given that the validity and reliability of some of these self-reported measures have not been established in the pediatric SCD population, interpretation of pain outcomes in this age group may be limited. ASCQ-Me subscales were used for predictor variables (eg, social functioning, emotional functioning, frequency of pain episodes) and the outcome variable (pain impact) in this study. Although ASCQ-Me is a measurement system that comprises multiple subscales that assess distinct constructs, it is important to acknowledge that shared method variance could potentially introduce bias into the study’s findings.

The interpretation of this study’s findings should also consider the variations in funding, resources, and availability of nonpharmacological treatments (eg, cognitive behavior therapy, acupuncture, and physical therapy) across different SCD centers. These potential differences in centers were not explored in this study. Moreover, some SCD centers participating in GRNDaD did not administer all 3 questionnaires (ASCQ-Me, BPI, and PROMIS Physical Function); thus, some participants lacked completed questionnaires. This limitation prevented direct comparisons between the 3 pain and pain-related models generated in this study. The presence of missing data is also a limitation that may introduce potential bias and affect the generalizability of the study’s findings. Lastly, limitations specific to multivariable regression modeling should also be considered. These include multicollinearity (ie, some of the predictor variables may be correlated) and overfitting, which would affect the generalizability of our models to the broader SCD population.

Despite these limitations, this study provides valuable insights into the factors influencing pain experiences and outcomes in individuals living with SCD. By conducting a secondary analysis of data from a large national, multicenter SCD registry, the study benefits from a substantial sample size, enhancing the generalizability of the findings. The multivariable modeling approach employed in the study is also a strength, as it allowed for the examination of various psychosocial and biological variables simultaneously. Our findings suggest that social and emotional functioning have a stronger association with pain impact in individuals with SCD compared to previously studied biological markers. This novel insight expands our understanding of the complex interplay between psychosocial and biological factors that influence pain perception and modulation in SCD, and thus, highlighting the importance of the biopsychosocial model in SCD pain. Importantly, given the observed increase in pain impact and pain severity during adolescence and young adulthood, this study brings attention to the need for focused and targeted pain interventions in this specific age group.

Insights gained from this study can inform future prospective studies that aim to establish the causal role of psychosocial variables in adverse pain outcomes in SCD. Future studies should also explore the relative contribution of pain-specific psychological variables (eg, pain coping and pain self-efficacy) and sociocultural variables (eg, social support) to pain impact and pain severity compared to biological modifiers SCD severity. By identifying modifiable biopsychosocial predictors of adverse pain outcomes in SCD, such studies can provide insight into the pathogenesis of chronic SCD pain, guide patient selection for trials of nonopioid interventions for SCD pain, and identify “high-risk” individuals for targeted preventive and symptomatic interventions early in their pain trajectory.