Impact of Race, Ethnicity, and Socioeconomic Status over Time on the Long-term Survival of Adolescent and Young Adult Hodgkin Lymphoma Survivors

There are approximately 90,000 new cancer diagnoses in adolescents and young adults (AYA; those aged 15–39 years) in the United States each year, and overall cancer incidence in this age group is rising (1). Recent data show improvements in cure rates, with 5-year overall survival (OS) rate of greater than 80% among AYA patients with cancer (2, 3). A small number of studies that have begun to investigate long-term mortality outcomes among AYA cancer survivors have consistently found that, compared with the general population, AYA survivors experience higher mortality rates that persist well into survivorship (4–6). As the AYA cancer survivor population continues to grow, more data are needed on long-term outcomes, specifically factors associated with inferior long-term survival.

Hodgkin lymphoma (HL) is one of the more prevalent cancers among AYA patients and has favorable 5-year survival of about 94% based on recent data from the Surveillance, Epidemiology, and End Results (SEER) database (2). In both childhood and AYA cancer survivor populations, those with HL have increased risks of long-term, noncancer mortality compared with other cancer types (7, 8), making investigation of demographic factors associated with long-term mortality especially important in this population. Socioeconomic status (SES), rurality, race, and ethnicity are known to impact 5-year survival in AYA patients with cancer, with low SES neighborhood, rural area residence, and black race associated with worse 5-year survival compared with high SES neighborhood metropolitan residence, and white race, respectively (9–13). Initial studies have noted racial disparities in 10-year survival of AYA patients with HL; however, more comprehensive long-term survivorship studies and those including recent diagnosis time periods are needed (14).

In the current large-scale retrospective analysis of SEER data, we sought to characterize long-term mortality patterns among 5-year survivors of AYA HL and determine the impact of race, SES, rurality, diagnosis age, sex, and HL stage on long-term survival in this population over time. In addition to its focus on disparities in long-term mortality, the current study utilizes a United States population-based registry. Much of the current long-term survivorship data in adolescent HL in the United States comes from the Childhood Cancer Survivorship Study, which is not a population-based cohort (15), while AYA survivorship data in the European population is drawn from population-based datasets such as the Teenage and Young Adult Cancer Survivor Study and national datasets including those from Nordic countries (16–18).

HL data extracted from SEER for the years 1975 to 2011 included data for 21,081 patients who were alive 5 years after initial diagnosis. The following exclusions were made: (i) missing race/origin code or classified as American Indian (n = 218); (ii) missing county-level statistics (n = 5); (iii) missing rural–urban continuum code (n = 226); (iv) those diagnosed prior to 1980 or after 2009 (n = 3,045); and (v) missing Ann Arbor Stage data (n = 1,688).

A county-level socioeconomic deprivation index (SES index) was defined based upon county-level variables previously defined by Truong and colleagues (19) County-level variables used in defining the SES index included the level of poverty (P) based upon percentage with income below the 200^th percentile of the poverty line, low educational attainment (E) per the percentage obtaining less than a high school education, crowding (C) per the percentage with room crowding with greater than one person per room, unemployment (U) per the percentage unemployed, levels of immigration (I) per the percentage of foreign-born, and language isolation (L) per the county percentage of language isolation. The variables P, E, C, U, I, L, were each standardized as the difference from the mean divided by the standard deviation. The county SES index was then calculated as (((P + E + C + U)/4) + ((I + L)/2))/2. Note that because this is defined as a socioeconomic deprivation index, higher values are indicative of greater deprivation.

Continuous variables were summarized by decade and overall (all decades pooled) as mean, standard deviation, median, minimum, and maximum. Discrete variables were summarized by decade and overall as count, percentage, mortality count, and mortality percentage. A time-to-event model was used to model the time (following 5-year overall survival) to death with relation to diagnosis decade (1980s, 1990s, 2000s), potential covariates including age at diagnosis (numeric years), rurality (numeric integer 0–8), race and origin (non-Hispanic white, Hispanic (all races), non-Hispanic black, non-Hispanic Asian or Pacific Islander), sex (female vs. male), Lymphoma Ann Arbor Stage (I–IV), lymphoma subtype recode per WHO 2008 [1(a)1.1 Lymphocyte-rich, 1(a)1.2 Mixed cellularity, 1(a)1.3 Lymphocyte-depleted, 1(a)2 Nodular sclerosis, 1(a)3 Classical Hodgkin lymphoma, NOS, 1(b) Nodular lymphocyte-predominant Hodgkin lymphoma], and SES index. Interactions with decade of diagnosis were also explored. A Cox proportional hazards model was considered (20), but due to extensive violations of the proportionality of hazards assumption, an accelerated failure time (AFT) model was selected instead; note that an advantage of utilizing accelerated failure time models is that, instead of hazard ratios, comparable interpretation is based upon intuitive survival time ratios (STR). The Weibull distribution was chosen as optimal for analysis due to lower Akaike Information Criterion (AIC) together with good fit of a Kaplan–Meier plot of model residuals to the model distribution. Based upon the AIC and beginning with a model including all variables and interactions with diagnosis decade, an exhaustive reverse process of model selection excluded first interactions followed by main effects variables where the exclusion yielded an improved model with lower AIC. The selection process retained the interactions between decade of diagnosis with Lymphoma Ann Arbor Stage and SES index, as well as the main effects variables.

The optimal model covariates resulting from the selection process included diagnosis decade, age at diagnosis, race and origin, sex, Lymphoma Ann Arbor Stage, lymphoma subtype, and SES index, with interactions of diagnosis decade with Lymphoma Ann Arbor Stage and SES index. Due to investigator interest, the final model included these variables and interactions, plus rurality, and the diagnosis decade interactions with sex and race and origin. The final time-to-event accelerated failure time model of overall survival time (following 5 years survival) to death included the variables diagnosis decade, age at diagnosis, race and origin, sex, Lymphoma Ann Arbor Stage, lymphoma subtype, rurality, and SES index, with interactions of diagnosis decade with sex, race and origin, Lymphoma Ann Arbor Stage, and SES index. Differences among discrete variable levels were assessed by contrasts Tukey-adjusted P values. Between-decade changes in discrete variable levels were estimated by contrasts with Hommel-adjusted P values. In the accelerated failure time model, follow-up time was taken into account when comparing survival between diagnosis decade.

An additional AFT model was implemented, which excluded decade interactions to allow overall assessment of effects across all decades for variables involved in interactions in the above model. This model was otherwise the same as that described above. Differences among discrete levels of these variables were pooled with those from the prior model, and Hommel-adjusted P values for the pooled estimates were derived from the pooled unadjusted P values. Reported P values reflect these adjustments.

Statistical analyses were performed using R statistical software (R Core Team, 2020, version 3.6.3; ref. 21). In all statistical tests, two-sided α = 0.05. Survival modeling was performed using the “survival” package (22, 23). Assessment of differences among discrete variable levels in the accelerated failure time model were estimated using the “emmeans” package (24); this includes adjusted means weighted proportionally to covariate marginal frequencies. Catseye plots were produced using the “catseyes” package (25, 26).

The characteristics of 15,899 5-year survivors of HL included in the analysis are shown in Table 1. Fifty percent of the cohort was female. Ten percent were black, 12% were Hispanic (all races), 4% were non-Hispanic Asian or Pacific Islander, and 74% were non-Hispanic white. The median age of diagnosis was 27 years and the median follow-up time (from 5-year survival) was 9.4 years (range, 0.1–28.9 years). Among survivors, 14% were diagnosed in the 1980s, 27% in the 1990s, and 58% in the 2000s. When considering stage, 20% of diagnoses were stage I, 49% were stage II, 19% were stage III, and 13% were stage IV. Overall mortality rate was 11%, and mortality rates (unadjusted for differential follow-up time) for those diagnosed in the 1980s, 1990s, and 2000s were 30%, 13%, and 5%, respectively (these raw mortality rates are unadjusted for differential follow-up times among the decades). The county-level SES index as well as the county-level variables, which informed the SES index, are summarized in Table 1, where the median SES index was −0.2. Accelerated failure time model results are summarized separately by variable below.

Overall, non-Hispanic black 5-year AYA HL survivors had significantly worse long-term survival compared with non-Hispanic white survivors, and while not reaching statistical significance, long-term survival of Hispanic survivors was also worse than that of non-Hispanic white survivors (Table 2). Non-Hispanic black survival was 66% that of non-Hispanic whites (STR: 0.66, P < 0.0001) and survival of the Hispanic (all races) patients was 78% that of non-Hispanic whites (STR: 0.78, P = 0.09), although not significant. There were no other significant differences in survival by race/ethnicity in the overall population. When stratifying by decade of diagnosis, racial/ethnic disparities in survival were significant for those non-Hispanic blacks diagnosed in the 2000s, attaining 54% (STR: 0.54, P ≤ 0.0001) of the survival attained by non-Hispanic whites; this trend was similar during the 1990s (STR: 0.66, P = 0.021). Hispanics of all races attained 69% (STR: 0.69, P = 0.20) of the survival attained by non-Hispanic whites diagnosed in the 2000s, although the trend lacked significance.

When assessing survival of 5-year AYA HL survivors over time, non-Hispanic blacks, Hispanics, and non-Hispanic whites each had improvements in survival for those diagnosed in the 2000s compared with those diagnosed in the 1980s (Table 2). Non-Hispanic blacks diagnosed in the 2000s had nearly twice the survival time as those diagnosed in the 1980s (STR: 1.89, P = 0.002), which was similar to Hispanics diagnosed in the 2000s compared with those diagnosed in the 1980s (STR: 1.91, P = 0.008). Non-Hispanic white 5-year survivors of AYA HL diagnosed in the 2000s had survival times of almost three times as long as those diagnosed in the 1980s (STR: 2.75, P < 0.0001). The survival disparity by race/ethnicity showed no evidence of change over time, with the non-Hispanic white to non-Hispanic black and Hispanic gap not significantly different between those diagnosed in the 2000s and those diagnosed in the 1980s (Table 2, P = 0.68 and 0.93, respectively).

Survival time decreased as the county socioeconomic deprivation index (SES index) increased with higher index values indicating higher levels of social deprivation and thus lower SES. Overall, each additional unit increase on the SES index was associated with an 8% reduction in survival time (STR: 0.92, P = 0.002, Table 3). When looking at SES disparities by decade, higher SES index was associated with decreased survival for AYA HL survivors diagnosed in the 1980s (STR: 0.761, P = 0.001). There were no significant differences in long-term survival by SES level for AYA HL survivors diagnosed in the 1990s or 2000s (P = 0.56 and P = 0.09, respectively). There was significant evidence of improving survival between the 1980s and 1990s (P = 0.003); however, significant evidence of this improvement was absent between the 1980s and 2000s (P = 0.32).

Age at diagnosis impacted long-term survival in 5-year AYA HL survivors (Table 3), with each additional year of age at diagnosis associated with a 4% reduction in survival time (STR: 0.964, P < 0.0001). There was no significant change in survival by rurality (P = 0.18) in our population (Table 3).

Female 5-year AYA HL survivors had significantly longer survival time than males, overall and when stratified by decade of diagnosis (Table 4). Overall, males attained 68% of females' survival time (STR: 0.68, P < 0.0001). Males diagnosed in the 1980s, 1990s, and 2000s attained 72% (STR: 0.72, P < 0.0001), 68% (STR: 0.68, P < 0.0001), and 60% (STR: 0.60, P < 0.0001), respectively, of the survival of females diagnosed in those same decades. Both female and male 5-year survivors of AYA HL had significantly improved survival over time. Female survivors diagnosed in the 2000s had nearly three times the survival time as those diagnosed in the 1980s (STR: 2.89, P < 0.0001). Male survivors diagnosed in the 2000s had over two times the survival time as those diagnosed in the 1980s (STR: 2.23, P < 0.0001). The survival disparity by sex did not change significantly over time (Table 4), with the female to male survival gap not significantly different between those diagnosed in the 2000s and those diagnosed in the 1980s (P = 0.43).

As expected, a higher stage at diagnosis was associated with worse long-term survival compared with those 5-year AYA HL survivors diagnosed at earlier stages (Table 5A). Focusing on overall changes, there was no difference in long-term survival among those diagnosed at stage II, compared with those diagnosed at stage I (P = 0.13). However, those diagnosed at stage III (STR: 0.68, P < 0.0001) and stage IV (STR: 0.59, P < 0.0001) had significantly worse survival than those diagnosed at stage I. In addition, 5-year AYA HL survivors diagnosed at stage III attained 81% the survival time attained by those diagnosed at stage II (STR: 0.81, P = 0.012), and those diagnosed at stage IV attained 70% the survival time of those diagnosed at stage II (STR: 0.70, P < 0.0001). There were no significant differences in long-term survival between 5-year AYA HL survivors diagnosed at stage IV and stage III. There was no evidence that the between-stage differences in survival changed between decades (Table 5B).

We found that racial, ethnic, and SES disparities in mortality persist for decades after diagnosis among 5-year survivors of HL diagnosed as AYAs. We also found that demographic and cancer-related factors, including age at diagnosis, sex, and HL stage impact long-term OS in this population. Specifically, black race or Hispanic ethnicity, higher county-level SES deprivation level, male sex, older age at diagnosis, and later HL stage were all associated with inferior long-term survival. Although previous studies have focused on early survival, the current study included only 5-year survivors of AYA HL and extends follow-time for up to 30 years. We found no evidence that these disparities, particularly racial/ethnic disparities, improved over time.

The findings in the current study are consistent with and expand on previous findings that race, ethnicity, and SES factors impact short-term survival in AYA patients with HL. We found that racial/ethnic disparities persist long-term among 5-year HL survivors, with both non-Hispanic black and Hispanic survivors having inferior long-term OS compared with non-Hispanic white survivors. This is consistent with previous data reporting decreases in both 5-year survival and longer-term survival among black AYA HL survivors, compared with white survivors; however, previous studies have found no differences in long-term survival between non-Hispanic whites and Hispanics (14, 27, 28). Recent data have found race/ethnicity based disparities among AYA HL survivors that likely contribute to the long-term survival disparities found in the current study. Compared with non-Hispanic white AYA HL survivors, non-Hispanic black survivors were more likely to have cardiovascular disease and both non-Hispanic black and Hispanic survivors were more likely to have endocrine disease, with any late effect doubling the risk of mortality in this population (29). Among adult cancer survivors, it has been shown that Hispanic and non-Hispanic black cancer survivors, compared with non-Hispanic white cancer survivors are less likely to report consistent and comprehensive follow-up care, such as seeing a specialist, adherence to preventive health services guidelines, and adherence to prescribed medications (30–34). In addition, compared with non-Hispanic white cancer survivors, non-Hispanic black and Hispanic cancer survivors are more likely to be obese, not meet physical activity guidelines, and not meet dietary guidelines (35–38). Taken together, these factors could place non-Hispanic black and Hispanic cancer survivors at increased risk of late mortality, compared with White cancer survivors and represent opportunities for focused intervention. Structural racism is also likely to be contributing to the persistent racial/ethnic disparities seen in the current study, particularly as it relates to health care access and SES disparities (39).

When assessing survival differences by decade of diagnosis, we found that there were no significant racial/ethnic differences in long-term survival for 5-year AYA HL survivors diagnosed in the 1980s. Overall, there has been significant improvement in long-term survival for AYA patients with HL diagnosed in more recent decades compared with those diagnosed in the 1980s (4). In this same timeframe, the toxicity profile of chemotherapeutic agents and radiation techniques has improved significantly (40). It is possible that the high long-term toxicity associated with HL treatments in the 1980s affected long-term mortality for those diagnosed in this decade to a greater extent than the aforementioned factors contributing to racial/ethnic differences in long-term survival, thus reducing racial/ethnic disparities in long-term mortality for those AYA HL survivors diagnosed in the 1980s. This study was also likely underpowered to assess racial and ethnic disparities within the cohort diagnosed in the 1980s.

The current study found that higher SES index (lower SES status) was associated with inferior survival compared with those AYA HL survivors with lower SES index (higher SES status). Previous studies have also shown that lower SES, compared with higher SES negative impacts survival at up to 15 years among AYA patients with HL (12). Low SES has been shown to impact access to care including findings of an increased likelihood of diagnosis delays and decreased use of subspecialty care, treatment, and survivorship care, which likely contribute to disparities in both short-term and long-term mortality (41–44). Specific to late effects of treatment, AYA cancer survivors living in low SES neighborhoods have a higher risk of cardiovascular disease compared with AYA cancer survivors residing in higher SES neighborhoods, which in turn, increases risk of long-term mortality (45). This risk is likely compounded for AYA HL survivors, who have higher risk of death after development of cardiovascular disease compared with most other AYA cancer types (45). In addition, education level on its own significantly impacts survivorship care, with lower education levels less likely to receive written survivorship care plans, and education less than high school associated with declining levels of medical care throughout survivorship (43, 46).

For patients with cancer, living in a rural or nonmetropolitan area compared with living in an urban or metropolitan area can affect both short- and long-term mortality through multiple pathways including increased symptom to treatment time interval, lower likelihood of being insured and receiving recommended therapies, decreased access to cancer specialty and support services, and higher prevalence of behavioral risk factors that increase the risk of mortality including smoking and lack of physical activity (47–52). Among AYA patients with cancer, data on differences in mortality by urban versus rural location are limited and mixed (53, 54). In the current study, we found that long-term survival was not impacted by rurality. However, rurality data were captured at the time of diagnosis, and the AYA age group tends to be highly mobile. According to the U.S. Census Bureau 2005–2010 data, overall 5-year moving rates were highest (65.5%) among those aged 25 to 29 years. Those aged 18 to 24 years had a 48% 5-year moving rate and those aged 30 to 44 years had a 45.5% moving rate. However, a large proportion of these moves were within county. Overall, among those aged 18 to 24, 25 to 29, and 30 to 44, 27.9%, 37.5%, and 27.6%, respectively, moved to a different location within the same county (55). Thus, it is plausible that some of the population included in the current study could have moved locations throughout the follow-up period, and the challenges of survivorship care in rural areas have not been captured.

The current study also found that higher stage at diagnosis, male sex, and older age at diagnosis, all known risk factors for early mortality, also increase the risk of long-term mortality among AYA HL survivors. Higher stage at diagnosis may require more intense chemotherapy, mediastinal radiation, and even allogeneic hematopoietic stem cell transplant, all increasing the risks of long-term morbidity and mortality (56). Older ages at diagnosis will naturally accrue age-related morbidities earlier in survivorship compared with those diagnosed at young ages. Similar to data from the childhood HL survivor population (57), dramatic sex-based differences in long-term survival among AYA HL survivors were seen, which may reflect this trend in the general population (58), as risks of major treatment-related causes of late morbidity in HL survivors including secondary neoplasms and cardiovascular disease do not differ by sex (59, 60). Despite similar incidence of cardiovascular disease in HL survivors, male survivors do have increased risk of cardiovascular mortality compared with female survivors, which likely contributes to the sex-based survival differences seen in the current study (61). Given that breast cancer is one of the most common secondary malignancies after HL treatment, it could be expected that women would have worse long-term survival, although studies have reported similar prevalence of second malignancies and mortality from solid tumors in male and female HL survivors (59, 61–63). Studies have also shown that there are no sex differences in receipt of recommended late-effects screening in AYA HL survivors (64); however, among AYA cancer survivors in general, male sex has been associated with less frequent engagement in survivorship care (65).

Importantly, the current study did not find any evidence of disparities in long-term survival in AYA HL survivors changing over time. This is an expected finding in age and stage at diagnosis disparities. However, it is a discouraging trend, particularly in race/ethnicity disparities, although this is consistent with recent data showing stable to worsening race/ethnicity disparities over time in 5-year survival among AYA patients with cancer (1, 10). In addition, AYA HL survivors are at increased risk of cardiac mortality, in which racial disparities in mortality have also increased over time (66, 67). Although the relationship between SES and race/ethnicity could not be fully resolved as SES was defined on the county level and race/ethnicity on the individual level, the findings in the current study are particularly striking given that race is an independent factor in the statistical model, thus minimizing the complex interplay between SES and race often found when assessing survival disparities. We also found that sex-based survival differences in AYA HL survivors did not change over time.

Although the SEER registries are widely considered as a definitive source of cancer data in the United States, some limitations come with using this dataset. The SEER dataset does contain some treatment information; however, data on specific chemotherapy and radiation regimens are not available. Thus, the extent to which treatment variables contributed to the disparities found in the current study is unknown. Rural populations are underrepresented in this dataset, which could have limited our ability to capture rural–urban disparities in long-term mortality of AYA HL survivors. Finally, these data describe sociodemographic variables associated with survival disparities, but cannot elucidate the reasons behind these differences. Socioeconomic data collected from SEER are not collected on an individual patient basis and SES status and patient location are not tracked over time, limiting conclusions that can be drawn. Census tract data in SEER are unavailable prior to the year 2000, and thus could not be utilized to study long-term survivors. We utilized these data to create a county level SES deprivation index, and although an imperfect substitute for individual level data, broadly, we found that county SES status impacts long-term mortality in AYA HL survivors. The extent to which SES impacts racial/ethnic survival disparities needs to be studied further and analyses including more refined SES data are needed to provide clarity on the impact of patient level SES status on long-term survival. A strength of this study is the large sample size that allows examination of each sociodemographic factor and analysis of changes in disparities over time. Follow-up time is also a strength.

The findings in the current study have demonstrated that race/ethnicity, SES, sex, age, and stage at diagnosis impact mortality outcomes in AYA HL 5-year survivors at up to 3 decades of follow-up. These data are unique in their focus on 5-year survivors and extend follow-up time beyond that previously assessed in this population. With improved cure rates for AYAs with HL, additional attention is needed to improve long-term outcomes for specific high-risk populations. Treatment-related modifications that have been implemented in the overall HL population, including decreasing utilization of mediastinal radiation and limiting exposure to toxic agents such as bleomycin, will likely continue to lead to improved long-term outcomes for AYAs; however, further studies are needed to understand why racial, ethnic, SES, and sex disparities persist years after cancer treatment has been completed. Identification of factors associated with long-term mortality among 5-year survivors of AYA HL may allow for targeted screening and follow-up by clinicians as well as the development of early interventions on known risk factors.

No disclosures were reported.

A.M. Berkman: Conceptualization, methodology, writing–original draft, writing–review and editing. C.R. Andersen: Formal analysis, methodology, writing–original draft, writing–review and editing. V. Puthenpura: Conceptualization, data curation, methodology, writing–original draft, writing–review and editing. J.A. Livingston: Conceptualization, writing–original draft, writing–review and editing. S. Ahmed: Writing–original draft, writing–review and editing. B. Cuglievan: Conceptualization, writing–original draft, writing–review and editing. M.A.T. Hildebrandt: Writing–original draft, writing–review and editing. M.E. Roth: Conceptualization, data curation, methodology, writing–original draft, writing–review and editing.

This work was supported by the National Cancer Institute at the National Institutes of Health (grant number P30 CA016672, M.E. Roth, J.A. Livingston, C.R. Anderson, and M.A.T. Hildebrandt) and (R38-HL143612, A.M. Berkman) and research support from the Archer Foundation and LyondellBasell (M.E. Roth, J.A. Livingston). The funders had no role in the design of the study, conduct of the study, analysis, interpretation of data, or decision to submit the manuscript for publication.

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.