Patient‐reported symptom burden in routine oncology care: Examining racial and ethnic disparities

Abstract Background Racial and ethnic disparities are well‐documented in cancer outcomes such as disease progression and survival, but less is known regarding potential disparities in symptom burden. Aims The goal of this retrospective study was to examine differences in symptom burden by race and ethnicity in a large sample of cancer patients. We hypothesized that racial and ethnic minority patients would report greater symptom burden than non‐Hispanic and White patients. Methods and results A total of 5798 cancer patients completed the Edmonton Symptom Assessment Scale—revised (ESAS‐r‐CSS) at least once as part of clinical care. Two indicators of symptom burden were evaluated: (1) total ESAS‐r‐CSS score (i.e., overall symptom burden) and (2) number of severe symptoms (i.e., severe symptomatology). For patients completing the ESAS‐r‐CSS on multiple occasions, the highest score for each indicator was used. Zero‐inflated negative binomial regression analyses were conducted, adjusting for other sociodemographic and clinical characteristics. Symptomology varied across race. Patients who self‐identified as Black reported higher symptom burden (p = .016) and were more likely to report severe symptoms (p < .001) than self‐identified White patients. Patients with “other” race were also more likely to report severe symptoms than White patients (p = .032), but reported similar total symptom burden (p = .315). Asian and Hispanic patients did not differ from White or non‐Hispanic patients on symptom burden (ps > .05). Conclusion This study describes racial disparities in patient‐reported symptom burden during routine oncology care, primarily observed in Black patients. Clinic‐based electronic symptom monitoring may be useful to detect high symptom burden, particularly in patients who self‐identify their race as Black or other. Future research is needed to reduce symptom burden in racially diverse cancer populations.


| INTRODUCTION
Racial and ethnic disparities in cancer outcomes are well-documented and evident at every stage of the cancer continuum, from prevention through active treatment and into survivorship. 1-3 Specific disparities related to race include lower cancer screening rates, higher incidence of certain cancers (e.g., multiple myeloma, colorectal, lung, cervical, and triple-negative breast cancer), increased perioperative mortality, and increased cancer-specific and overall mortality. [3][4][5][6] Reduction of health disparities is increasingly recognized as an important national goal. [7][8][9] One potential disparity that has received less attention is symptom burden.
Literature directly examining symptom burden among racial and ethnic groups is limited, typically focusing on single symptoms within specific cancer diagnoses in Black versus White patients. For example, Black or Hispanic women with breast cancer are more likely to report pain, skin irritations, and limitations in physical function when compared to those who are non-Hispanic and White. [10][11][12][13][14] Racial and ethnic disparities in symptom management have also been documented. Black patients report high levels of unmet needs in symptom management. [15][16][17] One study found that US-born Black patients and foreign-born Asian and Hispanic patients were up to 10.9% more likely to perceive an unmet supportive care need than White, US-born patients. 15 Further investigation is important to fully identify racial and ethnic disparities in symptom burden because under-or un-treated symptoms can lead to poor quality of life, higher rates of emergency department use, treatment non-compliance, end-of-life hospital admissions, and worse clinical outcomes. 15,[18][19][20] One way to evaluate symptom burden is through electronic clinic-based symptom assessments as part of routine clinical care.
Patient-reported outcomes (PROs) offer distinct information from provider-assessed adverse event reporting. For example, in a 2010 study of 1833 patient-health care provider dyads, providers significantly underestimated the presence of severe pain, fatigue, generalized weakness, anorexia, depression, constipation, poor sleep, dyspnea, nausea, vomiting, and diarrhea. 21 Another study comparing PROs with physician-assessed adverse events found that physicians under-reported severe treatment-related toxicities by up to 50%; under-reporting symptoms of any severity ranged up to 74%. 22 Conversely, recent studies indicate that clinic-based PRO assessment and symptom management results in better outcomes including improved patient-clinician communication, clinician awareness of patient symptoms, treatment decision making, healthcare utilization, patient satisfaction, quality of life, and survival. [23][24][25][26][27][28][29] The goal of the current, retrospective study was to examine potential racial and ethnic disparities in patient-reported symptom burden in adult oncology patients, controlling for other sociodemographic and clinical characteristics. It was hypothesized that patients self-identifying as a member of a racial or ethnic minority group (i.e., Black, Hispanic) would report higher total symptom burden and more severe symptoms compared to non-Hispanic and White patients.

| Participants and procedures
Patients presenting to the Moffitt Radiation Oncology or Supportive Care Medicine clinics completed the Edmonton Symptom Assessment Scale-revised (ESAS-r-CSS) 30 as part of routine clinical care. The strength of using a clinical dataset is that there is no recruitment bias.
Questionnaires were time-and date-stamped upon completion.
Patients were included in analyses if they were 18 years of age or older and had completed at least one symptom assessment. The study was approved by the Advarra Institutional Review Board. Ethnicity was categorized as Hispanic and non-Hispanic.

| Measures
Symptom burden: A modified version of the ESAS-r-CSS 30 was used to evaluate symptoms. The ESAS-r-CSS is a 12-item questionnaire that assesses the presence and severity of 12 core symptoms, including pain, tiredness, drowsiness, nausea, lack of appetite, shortness of breath, depression, anxiety, overall well-being, spiritual well-being, constipation, and difficulty sleeping. Patients rate each symptom on an 11-point Likert scale (i.e., 0 = none and 10 = worst possible) based on their symptoms at the time of questionnaire completion. Items were summed to create a total score (0-120). Higher scores indicate greater symptom burden. Individual symptoms were considered severe if they are rated 7 or greater. 31 The ESAS-r-CSS was administered on paper forms from January 2015 to January 2017 and via an electronic tablet thereafter. 31 All forms completed between January 2015 and June 2018 were included in analyses.

| Statistical analyses
Because the combined effect of several mild or moderate symptoms may be as burdensome as a severe symptom, symptom burden was assessed by two derived variables: (1) ratings of all symptoms were summed to provide a total symptom burden score and (2) the number of symptoms rated as severe was summed to separately to capture severe symptomatology.
For patients with data from multiple clinic visits, the highest score for each variable (total symptom burden and number of severe symptoms) was used. Scores for total symptom burden and number of severe symptoms could have been reported on the same clinic visit or different clinic visits.   Table 1. The majority of the sample was non-Hispanic (91%), White (86%), and male (54%). Patients ranged in age from 18 to  Note: Percentages calculated from available data. 97, with an average age of 64 (SD = 13). The three most common cancer diagnoses were lung (18%), breast (18%), and male genitourinary (16%). Slightly more than half of the sample had active cancer at the time of assessment (53%). A majority of those with known marital status were married (69%). However, marital status was missing for 16% of patients. Due to this high proportion of missing data, marital status was not considered for inclusion in the ZINB models.
Patients completed a total of 19 670 individual surveys. Fewer than 3% of patients were missing any ESAS information. The median and range of surveys contributed by each patient was 2 (1-29). The highest overall symptom burden and highest severe symptom burden were retained for each of the 5798 patients. Patients' average worst overall symptom burden score was 31.2 (SD = 24.5) out of a possible score of 120. Responses ranged from 0 to 116, with a median of 24.
An overall worst symptom burden of zero was reported by 356 patients (6.2%). Participants' average highest number of severe symptoms was 2.0 (SD = 2.5) out of a possible score of 12. Responses ranged from no severe symptoms to all 12 symptoms rated as severe, Note: In general, larger (positive) estimates indicate greater expected total symptom burden scores. Ratios for categorical variables report the expected total symptom burden among those not already predicted to have zero burden for the comparison group divided by the expected total symptom burden for the reference group. For age, the ratio reports the expected multiplicative change in total symptom burden associated with each additional year of age. Note: For categorical covariates, estimates correspond to the expected change in the logarithm of the symptom score for each comparison group compared to the logarithm of the symptom score for the reference group (e.g., Black/African/American vs. White). Positive estimates (and ratios exceeding one) correspond to higher symptom scores for the comparison group, while negative estimates (and ratios below one) correspond to higher symptom scores for the reference group. For continuous covariates (e.g., age), estimates report the expected change in logarithm of the symptom score per unit change in the covariate.

| Modeling overall symptom burden
The first step of the ZINB models was to identify predictors of zero symptom burden using multivariable logistic regression models.
Results of the backward selection logistic regression analysis indicated that being male, not having active disease, and location of the primary cancer site were variables associated with higher odds of having a symptom burden of zero (Table S1).

| Modeling severe symptomatology
Analyses were conducted with the same procedures described above to identify predictors of zero severe symptoms using multivariable logistic regression models. Results of the backward selection logisticregression analysis indicated that older age (p = .0364), male sex (p = . 0311), absence of active disease (p < .0001), and primary cancer site (p < .0001) were associated with the probability of no severe symptoms (Table S2) Note: Estimates correspond to the expected change in log-odds of zero symptom score. Larger (positive) estimates indicate greater odds of zero score. Odds ratios are equal to the odds of having zero total symptom burden in the comparison group divided by the odds of having zero total symptom burden in the reference group. Note: For categorical covariates, estimates correspond to the expected change in the logarithm of the symptom score for each comparison group compared to the logarithm of the symptom score for the reference group (e.g., Black/African/American vs. White). Positive estimates (and ratios exceeding one) correspond to higher symptom scores for the comparison group, while negative estimates (and ratios below one) correspond to higher symptom scores for the reference group. For continuous covariates (e.g., age), estimates report the expected change in logarithm of the symptom score per unit change in the covariate. primary cancer site in the model for total symptom score. In Table 5, the parameter for bone cancer has a wide confidence interval, likely due to the small sample of patients with bone cancer. Paralleling the findings for total symptom burden, the zero-inflated negative binomial model for severe symptom burden provided sufficiently more information compared to a model without zero inflation to justify the use of the zero-inflated model (p < .0001).
T A B L E 4 Zero-inflated negative binomial regression assessing racial and ethnic differences in the maximum number of severe symptoms Note: In general, larger (positive) estimates and odds ratios exceeding unity indicate greater expected total symptom burden scores. Ratios for categorical variables report the expected number of severe symptoms among those not already predicted to have zero severe symptoms for the comparison group divided by the expected number of severe symptoms for the reference group. For age, the logarithm of the ratio reports the expected change in number of severe symptoms associated with an additional year of age. Note: For categorical covariates, estimates correspond to the expected change in the logarithm of the symptom score for each comparison group compared to the logarithm of the symptom score for the reference group (e.g., Black/African/American vs. White). Positive estimates (and ratios exceeding one) correspond to higher symptom scores for the comparison group, while negative estimates (and ratios below one) correspond to higher symptom scores for the reference group. For continuous covariates (e.g., age), estimates report the expected change in logarithm of the symptom score per unit change in the covariate. Note: All bolded values are statistically significant; see table for specific p-values.
The negative binomial component of the ZINB model facilitates an analysis of the expected number of severe symptoms for patients predicted to have at least one severe symptom. In addition to race and ethnicity, the final negative binomial model included age, sex, primary cancer site, and cancer disease status. Results of the ZINB regression model indicated that race was associated with the count of severe symptoms (p = .0002). Specifically, the expected number of severe symptoms for patients predicted to have at least one severe symptom was 24% higher for Blacks, 21% lower for Asian patients, and 20% higher for those of "other" race compared to White patients.
No statistically detectable differences in severe symptoms were observed between Hispanic and non-Hispanic patients (p = .43).
Females predicted to have nonzero severe burden reported 20% higher expected counts of severe symptoms than males (p < .0001).
Patients without active disease reported 20% lower expected counts of severe symptoms than did patients with active disease (p < .0001).

| DISCUSSION
The goal of this study was to examine potential racial and ethnic disparities in patient-reported symptom burden and severe symptomatology in cancer patients. We hypothesized that patients selfidentifying as a member of a racial or ethnic minority group (i.e., Black, Hispanic, Asian) would report higher total symptom burden and more severe symptoms compared to non-Hispanic and White patients. Our hypotheses were partially supported, as analyses revealed disparities in higher overall symptom burden and more severe symptoms reported by Black patients as compared to their White counterparts.
These results suggest that there is an unmet need for symptom management in patients who self-identify as Black, and that supportive care interventions should be explored in order to reduce the high level of severe symptoms and impact of symptom burden in this patient group.
T A B L E 5 Analysis of maximum likelihood zero inflation parameter estimates in the maximum number of severe symptoms Note: Estimates correspond to the expected change in log-odds of severe symptoms. Larger (positive) estimates indicate greater expected odds of reporting zero severe symptoms. Odds ratios are equal to the odds of having zero severe symptoms in the comparison group divided by the odds of having zero total symptom burden in the reference group. Note: For categorical covariates, estimates correspond to the expected change in the logarithm of the symptom score for each comparison group compared to the logarithm of the symptom score for the reference group (e.g., Black/African/American vs. White). Positive estimates (and ratios exceeding one) correspond to higher symptom scores for the comparison group, while negative estimates (and ratios below one) correspond to higher symptom scores for the reference group. For continuous covariates (e.g., age), estimates report the expected change in logarithm of the symptom score per unit change in the covariate. Note: All bolded values are statistically significant; see table for specific p-values.
We anticipated that Hispanic patients would report worse symptom burden and more severe symptoms than non-Hispanic patients, in part due to previous findings that Hispanic cancer patients are at risk of higher symptom burden, 34 worse psychological distress, 35 40 Prior research indicates that racially discordant interactions (typically where the patient is a racial minority and the physician is White) are common and perceived by the patient as less positive and productive. 41 Racially discordant appointments are also shorter in length, and the discussion is typically more physiciandominated and less patient-centered. 42 These factors likely create a barrier in patients' comfort reporting their symptoms to their provider, which may contribute to disparities in effective symptom management for racial minorities. This healthcare system-level risk factor may be particularly important because it may be more easily modified than some individual-and system-level factors. While research suggests that minority patients may have a better experience when treated by racially concordant providers, 43  symptom burden, number of severe symptoms); routine collection of data as part of clinical care that includes all patients; and an integrated electronic medical record system that allowed ESAS-r-CSS assessments to be linked with sociodemographic and clinical characteristics.