Disparities in the Early Adoption of Chemoimmunotherapy for Diffuse Large B-cell Lymphoma in the United States

With an estimated 66,360 new cases diagnosed in the United States in 2011, non-Hodgkin lymphoma (NHL) is the sixth most common cancer (1). Diffuse large B-cell lymphoma (DLBCL) is the most commonly occurring subtype of NHL in the United States, comprising approximately one-third of all adult lymphomas. The natural history of DLBCL is aggressive, with a median survival of less than one year in untreated patients (2). The combination chemotherapy regimen of cyclophosphamide, doxorubicin, vincristine, and prednisone (CHOP) remained the standard therapy for DLBCL for decades following its development in the 1970s (3, 4). Even when compared with more intensive chemotherapy regimens, the standard CHOP regimen produced similar survival outcomes and was better tolerated (3).

In 1997, rituximab, the first monoclonal antibody used as an anticancer therapy, was approved for use by the U.S. Food and Drug Administration (FDA) for follicular lymphoma, and this immunotherapy was soon applied to DLBCL and other B-cell NHLs (5, 6). In 2002, a randomized clinical trial in patients >60 years of age with advanced-stage DLBCL showed that when rituximab was added to standard CHOP chemotherapy, the complete response rate improved from 63% (CHOP alone) to 76% (P = 0.005), and 2-year overall survival (OS) improved from 57% to 70% (P = 0.007; ref. 7). Follow-up data from this trial and other randomized, controlled clinical trials confirmed the benefits of combined chemoimmunotherapy for DLBCL (8–10). For patients with limited-stage DLBCL chemotherapy given with rituximab for 3 cycles followed by radiation has produced 7-year OS rates of up to 95% (11, 12) and has become the acknowledged standard of care for this group as well (13).

However, it remains unclear what the timing of adoption of chemoimmunotherapy as the standard for DLBCL has been in the United States and what factors have influenced who received this combination when it was first adopted. To address these issues we sought to (i) describe the clinical and demographic features of patients with DLBCL who received combination chemoimmunotherapy during the time period of its initial adoption, (ii) assess the differences between patients with DLBCL who received chemoimmunotherapy and those who did not, and (iii) examine time trends in the use of chemoimmunotherapy for patients with DLBCL.

Data from the National Cancer Data Base (NCDB), a nationwide, hospital-based cancer registry jointly sponsored by the American Cancer Society and the American College of Surgeons, were used to examine chemoimmunotherapy use. The NCDB contains approximately 20 million records from more than 1,400 Commission-on-Cancer (CoC) approved hospital-based cancer programs in the United States and Puerto Rico. Whereas the NCDB accrues cancer cases arising in the hospital-based setting, this database collects patient-level data from encounters that occur in both the inpatient and outpatient setting. Because cancer diagnoses are determined by pathologists and pathology departments that are largely hospital-based, diagnoses of DLBCL and treatment of DLBCL that occurs in the outpatient setting would be captured in the NCDB as long as the pathologist rendering the diagnoses carries out services in a CoC hospital. Each record contains standardized data elements on patient demographics, tumor characteristics (including stage and histopathology), and first course of therapy. Approximately 75% of all newly diagnosed cases of cancer in the United States are captured at the institutional level and includes inpatient and outpatient data reported to the NCDB. The NCDB also contains information on patient insurance status, county of residence, facility type in which patients were treated, and an encrypted facility identifier. Because no patient, provider, or hospital identifiers were examined and no protected health information was reviewed, Institutional Review Board approval was not required for this study.

Patients were eligible for study inclusion if they were aged 18 years or older, diagnosed with International Classification of Diseases for Oncology (ICD-O; ref. 14) codes 9679 or 9680 between January 1, 2001, and December 31, 2004, and received all or part of their first course of treatment at the reporting facility in the inpatient or outpatient setting. All data refer to neoplasms with malignant behavior. Data were abstracted using coding guidelines documented in the Facility Oncology Registry Data Standards (FORDS) manual.

Among the 51,207 patients identified with DLBCL, 38,002 met all eligibility criteria (Supplementary Fig. S1). Patients were excluded if their age was <18 or >101 years, or if there were missing data on age, gender, race/ethnicity, insurance, region of residence, chemotherapy, or immunotherapy. Missing data on race was the most common reason for exclusion (n = 4,527, 8.84%). Non-Medicare/Medicaid government-funded plans comprised 0.3% of the population and were excluded from analyses because they provided insufficient data to be separately analyzed.

Administration of chemotherapy, immunotherapy, and radiation therapy was captured using the FORDS. To examine recent treatment practice patterns that were in accordance with state-of-the-art available data in 2002, we examined factors associated with chemotherapy plus immunotherapy versus chemotherapy alone. The independent variables included in this analysis were drawn from 4 sources: patient-level demographics, clinical characteristics, facility-level variables, and area-level information. Patient-level demographic variables included age at diagnosis, gender, race/ethnicity, and primary payer or insurance type at diagnosis. Clinical characteristics captured at the patient level included lymphoma stage, primary site, and year of diagnosis. Treatment-level or facility-level characteristics included the volume of lymphoma patients at the treatment facility-by-facility type. Area-level characteristics were based on the patient's residence and included census region and education level in patient's zip code. Because all area-level measures of socioeconomic status (SES) were highly correlated, educational level was selected as a single marker of SES.

Three types of treatment facilities were described according to the classification system used by the College of Surgeons Commission on Cancer's approvals program; these are community cancer programs, comprehensive community cancer programs, and teaching/research centers. Community hospitals treat at least 300 cancer cases a year and have a full range of services for cancer care, but patients need referral for portions of their treatment. Comprehensive community cancer centers are facilities that offer the same range of services as the community hospitals but have at least 750 annual cancer cases and conduct weekly cancer conferences. Teaching/research facilities differ from community cancer facilities in that the teaching/research facilities have residency programs and ongoing cancer research. Twenty-nine of the 39 National Cancer Institute designated Comprehensive Cancer Programs participate in the CoC approvals program and are included among teaching/research facilities in this study.

Treatment volume for each facility was calculated by counting all DLBCL cases diagnosed during 2001–2004. The facilities were then divided into low or high volume based on the median by facility type, which was 13 cases over the study period for community hospitals, 37 for comprehensive community cancer centers, and 55 for teaching/research facilities. Patient residence was based on the reported state of residence at diagnosis and categorized as West, Midwest, Northeast, or South. Area-based indicator of patient education were derived at the zip code level from 2000 U.S. Census data and included as quartiles of the observed distribution in the general U.S. population. The proportion of the population in a patient's zip code of residence who did not have a high school diploma was stratified as ≥29%, 20%–28.9%, 14%–19.9%, <14%, or missing.

To evaluate how representative our cohort of patients was to a population-based sample, we compared our study cohort to Surveillance, Epidemiology, and End Results 17 (SEER) cases diagnosed with DLBCL (ICD-0 codes 9679 and 9680) between 2001 and 2004. To assess the capture of immunotherapy and chemotherapy in the NCDB, we compared rates of chemotherapy and immunotherapy among NCDB patients ≥65 years to patients in a linked SEER-Medicare dataset. SEER-Medicare was projected to have more complete data on both chemotherapy and immunotherapy because these are billed separately, and rituximab has a specific Healthcare Common Procedure Coding System code.

Analyses were carried out with SAS software (version 9.2; SAS Statistical Institute). χ² tests were used to analyze the relationship between race, other covariates, and the outcome variables of interest. Because all other variables examined were statistically significant in univariate analyses, multivariable log binomial models were used to generate risk ratio estimates (α = 0.05). Multivariable log binomial models were relied upon considering the common outcome in our study (15). Interactions between race and all other covariates were tested by comparing -2 Log Likelihood χ² values from Chunk Tests. Time trends in the use of chemoimmunotherapy for patients with DLBCL were tracked from 1998 to 2004 and compared by race. The Cochran Armitage test for trend was used to examine time trends in the use of chemoimmunotherapy for DLBCL.

Patients with DLBCL in NCDB and SEER were similar in terms of their age, gender, year of diagnosis, and primary site of involvement (Table 1). Fewer DLBCL patients in the NCDB cohort had unknown stage. A greater percentage of DLBCL cases in the NCDB cohort were from the South and Northeast regions, and relatively fewer cases were from the West. No data are available in SEER about the primary payer, facility characteristics, use of chemoimmunotherapy, or area-level measures of SES during this time period. Male patients that were uninsured or with Medicaid insurance presented at a younger age compared with patients with private insurance [median age 48, 47 years vs. 55 (P ≤ 0.0001)] and more commonly presented with advanced stage disease [stage III/IV 46.1%, 51.9% vs. 39.5% (P ≤ 0.0001)]. Similarly, females that were uninsured or with Medicaid insurance were younger [median age 54, 52 years vs. 56 (P ≤ 0.0001)] and more commonly presented with advanced stage disease [38.6%, 42.9% vs. 36.8% (P ≤ 0.0001)]. In accordance with our prior findings using SEER data (16), black patients presented with DLBCL a decade younger and more commonly had advanced stage disease. In NCDB, black and Hispanic patients more commonly were uninsured or had Medicaid insurance and were more commonly treated at a low volume comprehensive cancer facility (Table 2).

Within the NCDB, 10,234 patients were identified who received combination chemoimmunotherapy and 19,078 who received chemotherapy alone. There were statistically significant differences between these groups in terms of stage at diagnosis, diagnosis year, race, primary payer, region of the United States, facility type, and measures of SES (Table 3). The probability of receiving chemotherapy among DLBCL patients age 65 and older was similar in NCDB (71.1%) and SEER-Medicare (68.1%). Although, the proportion of patients who received chemotherapy plus immunotherapy was lower among NCDB patients (24.8%) compared with SEER-Medicare (46.3%), the under ascertainment of immunotherapy in NCDB seemed to be nondifferential with respect to race. The ratio of chemoimmunotherapy use among black relative to white patients was similar in NCDB (0.80) and SEER-Medicare (0.77). In addition, the proportion of patients receiving no treatment was similar between NCDB patients ≥65 years (19.9%) and SEER-Medicare patients (21%).

In the multivariable log binomial models, year of diagnosis, stage, race, age >60 years, region, area-level educational status, and facility characteristics were significant predictors of receiving combination chemoimmunotherapy versus chemotherapy alone and were significant predictors of receiving any chemotherapy (alone or with immunotherapy) versus no chemotherapy (Table 4). There was no association between insurance status and receipt of chemoimmunotherapy versus chemotherapy alone; however, insurance was associated with receipt of chemotherapy versus no chemotherapy. Patients who were black [relative risk (RR), 0.83; 95% confidence interval (CI), 0.78–0.89], >60 years (RR, 0.94; 95% CI, 0.90–0.98), had limited stage disease (RR, 0.89; 95% CI, 0.86–0.92), or missing staging information (RR, 0.54; 95% CI, 0.50–0.58) were diagnosed in 2001–2002, or were treated at a facility other than a high volume teaching/research facility were less likely to receive chemoimmunotherapy. Patients who were black (RR, 1.14; 95% CI, 1.05–1.25), Hispanic (RR, 1.23; 95% CI, 1.13–1.35), or >60 years (RR, 1.39; 95% CI, 1.30–1.50) were more likely to receive no form of treatment following their diagnosis. In addition, patients without insurance or insurance other than private were more likely to receive no treatment, as were patients treated at low volume facilities.

From 1998 to 2004 the proportion of patients who received chemoimmunotherapy in NCDB increased dramatically from 0.01% to 32.6% (Cochran Armitage Z = −78.48, P < 0.0001). The increase in the use of chemoimmunotherapy over this time period occurred for patients across all racial categories (Fig. 1A), but disparities existed throughout this period. In 2004, 23.6% of black patients and 33.2% of white patients received chemoimmunotherapy as the initial treatment for DLBCL. Similarly, in 2004 26.6% of Medicaid patients and 27.3% of uninsured patients received chemoimmunotherapy, compared with 35.9% of privately insured patients (Fig. 1B).

Although lymphoma represents 5% of all cancers in the United States, it is estimated that $4.6 billion per year is spent in the United States on treatment for lymphoma (17). This high cost of lymphoma care fortunately has been associated with improvements in outcomes. The CHOP regimen has been the foundation of therapy for DLBCL for several decades, despite attempts to improve outcomes with more intensive treatments (3). When the Groupe d'Etude de Lymphome d'Adultes (GELA) reported the first randomized controlled trial showing the benefit of adding rituximab to CHOP chemotherapy for the treatment of patients ≥60 years of age with newly diagnosed DLBCL, the standard of care began to change. Additional data from the MabThera International Trial showed that patients ≤60 years chemoimmunotherapy experienced improved 3-year event-free survival from 59% (chemotherapy alone) to 79% (log-rank P < 0.0001), and increased 3-year OS from 83% to 93% (log-rank P = 0.0001; ref. 18). The results from the GELA trial in older patients were confirmed by a U.S. Intergroup trial (10), the German Lymphoma Study Group RICOVER-60 trial (19, 20), and additional follow-up of the GELA study (8). Furthermore, a population-based study comparing adult patients in British Columbia with DLBCL treated when the province standard therapy was CHOP or CHOP-like chemotherapy to patients treated under an updated policy that recommended chemotherapy with rituximab as standard of care showed improved progression-free survival and OS among patients treated under the updated policy (21). Although the final publication of some of these results occurred following our study period, substantial data were presented earlier at peer-reviewed international meetings describing the benefits of chemoimmunotherapy (5, 6, 22).

Following the initial FDA approval of rituximab in 1997, the proportion of patients who received chemoimmunotherapy for DLBCL increased dramatically from 1998 to 2004. In this study, we found that although increase in the use of chemoimmunotherapy occurred for patients across all racial and SES categories, uninsured, Medicaid, black, and lower SES patients were less likely to receive chemoimmunotherapy in the 2001–2004 time period. In multivariable models, age >60 years, stage, race, region, SES, and facility characteristics remained significant predictors of receiving chemoimmunotherapy, but insurance status did not. However, insurance status was likely not a significant predictor of chemoimmunotherapy because uninsured patients were less likely to receive any treatment at all, making chemotherapy a poor comparator. The reduced use of chemoimmunotherapy in older DLBCL patients is particularly interesting, given that the benefits of chemoimmunotherapy were first shown in patients >60 years of age (7). Follow-up national cohort studies are needed to determine whether this was a time-limited phenomenon or may reflect a bias toward limiting therapy in older individuals. Racial and SES disparities in the use of novel technologies or therapies for cancer patients have previously been shown (23). Further studies are necessary to disentangle the interacting effects of race- and individual-level SES in national cohort studies of lymphoma treatments and outcomes and to assess the influence that patient-level variables like social support, emotional support, and informational support have on lymphoma treatment selection (24). Because individual-level SES is not available in the NCDB, we were not able to examine the complex relationship between race and SES on lymphoma treatments. However, our analysis suggests that even when controlling for other covariates, age, race, and area-level SES remain important predictors of who received chemoimmunotherapy for DLBCL. Unfortunately, additional follow-up of these trends is not possible using the NCDB because coding of rituximab immunotherapy changed following this period.

In addition, we found that patients who received treatment at a high lymphoma volume teaching/research center or comprehensive cancer center were more likely to receive chemoimmunotherapy. High-volume research and teaching facilities seemed to be early adopters of this innovation and may be able to more promptly apply treatment advances to individual patients. Understanding factors that influence the adoption and use of innovative treatment strategies (25) at these facilities may provide insight into measures that can be applied to speed the diffusion of innovative treatments to other segments of the U.S. population.

It is important to note some limitations of this study. Because registry data do not contain the specific type of chemotherapy or immunotherapy administered, we relied upon general codes to indicate whether chemotherapy and/or immunotherapy were administered. Although rituximab was the major form of immunotherapy administered during this time, it is possible that other forms of immunotherapy (e.g., IFN) could have been coded as well. However, the use of these therapies in the United States during this time period was limited. Moreover, area-level measures of SES were available in this dataset but patient-level measures of SES were not. In other studies in which patient-level SES were collected with area-level SES, the latter seemed to underestimate the effect of SES (26). Because cancer registry data lack information on all relevant clinical data such as International Prognostic Index (IPI) score (27), our analyses could not examine the impact of these factors on chemoimmunotherapy use. However, where possible we did integrate components of the IPI (age >60, stage III/IV) as covariates in our regression models. Our study is also subject to the effects of potential misclassification and under reporting previously described in cancer registry studies, including for race/ethnicity and treatment. However, CoC-approved facilities are thought to have more complete data on treatment (28). Ascertainment of patients for this study was based on a diagnosis of DLBCL being rendered at a CoC hospital facility, including patients who received care in both the inpatient and outpatient settings. This would reduce the possibility of selection bias because of more black patients being treated as inpatients because of advanced disease compared with white patients.

When we compared patients in our cohort aged >65 to SEER-Medicare, the rates of chemotherapy use and no treatment were similar. However, the rate of chemoimmunotherapy among NCDB patients was lower than that observed among SEER-Medicare patients, which may reflect improved receipt of chemoimmunotherapy among the Medicare population and/or an underascertainment of immunotherapy in NCDB. However, the underascertainment of immunotherapy did not vary by race. Moreover, because black patients less commonly present with DLBCL over the age of 65 years, SEER-Medicare linked data may be a less appropriate dataset for examining racial disparities in DLBCL treatment, further accentuating the value of our analysis using the NCDB.

Despite these limitations, this study has several strengths. First, the NCDB provided a large national cohort of patients with validated information on demographic and clinical characteristics at diagnosis. Moreover, it provides data on insurance status and area-level SES, allowing discrimination between the effects of race, insurance status, and SES on treatment administration. Finally, this study used a cohort of patients from over 1,350 institutions across the United States over a 4-year time period surrounding the demonstration that rituximab with chemotherapy improved outcomes over chemotherapy alone. This allowed us to examine the diffusion of this innovative treatment strategy across segments of the U.S. DLBCL population and to explore factors that predict who received chemoimmunotherapy.

Other studies have also showed that although advances in therapy have produced improvements in survival for patients with NHL, disparities exist in treatment and outcomes in the United States (29–31). In particular, a SEER-Medicare analysis by Wang and colleagues showed that older African American patients with NHL were less likely to receive chemotherapy than Caucasian patients (31). Although lower SES was predictive of all-cause and NHL-specific mortality, after controlling for differences in treatment, comorbidity, and socioeconomic status, race was not. In this study by Wang and colleagues, 72% of African Americans resided in areas with the poorest quartile of SES as compared with 22% of Caucasians, making it difficult to separate the effects of race and SES on treatment selection and outcome. More importantly, the study did not distinguish among NHL subtypes, rendering their findings about treatment selection and survival difficult to interpret as these vary markedly by lymphoma subtype. In our study, black patients were younger (median age 53 vs. 70 years), more likely to present with stage III/IV disease (44.5% vs. 40.9%), more likely to be uninsured (9.5% vs. 2.5%) or Medicaid insured (17.3% vs. 3.4%), and more likely to reside in a zip code where ≥29% of the population had no high school diploma (38.1% vs. 11.6%) when compared with white patients (all P < 0.0001). However, there was no effect modification between SES and insurance on race after controlling for demographic (gender, age), clinical (diagnosis year, primary site, and stage), and facility-level factors. Moreover, our study focused solely on patients with DLBCL eliminating the effect that known racial differences in NHL prevalence would have on our findings (16, 32).

These results support our prior findings that black patients with DLBCL in the United States present at a younger age and with more advanced stage disease when compared with other racial/ethnic groups (16). Among 37,009 DLBCL cases diagnosed from 1992 to 2005 in the SEER registry, 65% of black patients compared with 37% of whites presented at age ≤60 years (median age 51 vs. 68, P < 0.001). Moreover, 54% of black compared with 47% of white patients presented with stage III/IV disease (P = 0.05) and 5-year survival rates were 38% versus 46% (P = 0.02). These differences in presentation did not arise through an association with HIV (33), which was also shown in the NCDB (data not shown).

Do disparities in treatment outcomes result from inequalities in the use of standard therapies for lymphoma? Although a number variables may influence disparities in treatment outcomes, including patient-related, provider-related, healthcare system-related, and societal factors (34–37), prior studies investigating healthcare disparities in cancer patients indicate that equal treatment yields equal outcomes (38–41). For lymphoma patients limited data exist with regard to disparities in treatment usage that are linked to treatment outcomes. A recent SEER analysis comparing survival trends among patients with DLCBL from 1973 to 2004 showed improved median OS in the era of immunotherapy [2000–2004 median OS 47 months, (P = 0.005)]; however this benefit was not maintained across race, with white patients having significantly better outcomes (47 months vs. 29 months; P = 0.001; ref. 42). Moreover, an Emory University cohort study that examined 361 white and 123 black DLBCL patients indicated that racial differences in DLBCL outcomes occur even when the same treatment is given to black and white patients (43). Although there were no racial differences in the use of R-CHOP therapy, white race predicted for improved 2-year OS (odds ratio, 1.77; 95% CI, 1.15–2.73). Because there are known biologic subgroups of DLBCL that are associated with differences in treatment response and survival (44), carrying out meaningful DLBCL disparities research in the future will require collection of biologic specimens to ascertain the interactions between, race, SES, treatment disparities, and outcomes.

Effective lymphoma care involves the provision of appropriate and beneficial services to cancer patients based on scientific knowledge. Quality of care needs to be defined and studied to ensure that patients with DLBCL and other cancers receive the best available standard of care. Although our study showed that the use of chemoimmunotherapy rose during the period immediately following proof of its benefit, improving outcomes for patients with lymphoma in the United States will require increased attention on efforts to extend the benefits of proven advances in therapy to all segments of the population.

C.R. Flowers has served as a consultant for Spectrum, Celgene, Optum Rx, Seattle Genetics, Allos, Genentech/Roche (unpaid), and Millennium/Takeda (unpaid). He also leads research studies that are supported by Spectrum, Novartis, Millennium/Takeda, and Gilead.

Conception and design: C.R. Flowers, A.Y. Chen, O.W. Brawley

Development of methodology: C.R. Flowers, S.A. Fedewa, A.Y. Chen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.M. Ward

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.R. Flowers, S.A. Fedewa, A.Y. Chen, J. Lipscomb, O.W. Brawley, E.M. Ward

Writing, review, and/or revision of the manuscript: C.R. Flowers, S.A. Fedewa, A.Y. Chen, L.J. Nastoupil, J. Lipscomb, E.M. Ward

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.A. Fedewa

Study supervision: C.R. Flowers

The authors thank Dr. Katherine Virgo and Dr. John Bian for analyzing the rates of chemotherapy and immunotherapy in SEER-Medicare data.

This work was supported by Dr. Flowers' Georgia Cancer Coalition Distinguished Scientist Award and American Society of Hematology Amos Medical Faculty Development Award. Research reported in this publication was also supported by the National Cancer Institute of the National Institutes of Health under Dr. Flowers' Award Number R21CA158686. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.