Neighborhood and Individual Socioeconomic Disadvantage and Survival Among Patients With Nonmetastatic Common Cancers

1Socioeconomic status (SES) is a complex concept describing the state of income, wealth, education, occupation, and living conditions for both individuals and their neighborhood or social network., Individual SES has consistently been reported to be an important risk factor for worse survival across a range of cancer types.,,,,,,,,,,,,, However, investigators often focus on the associations of neighborhood SES with cancer survival in large national cancer databases in which individual-level SES is not accessible,; thus, individual-level SES is underexamined in these national databases. Owing to systemic inequalities in opportunity, development, and built environment, disadvantaged neighborhood SES may result in worse cancer outcomes via multiple pathways. For example, in deprived neighborhoods, lack of physicians and health care resources, inefficacious referral systems, low trust in health care providers, poor social support networks, and barriers to travel for initial and follow-up care may together negatively impact survival after cancer diagnosis.,,,,,,,

2Despite the understanding that both individual and neighborhood SES influence cancer outcomes, important knowledge gaps exist. First, it is unclear whether neighborhood SES is independently associated with cancer outcomes after accounting for individual-level SES, or whether there is a synergistic association between these 2 factors., To our knowledge, the interaction of individual-level and neighborhood-level SES with cancer survival has not been previously compared. Second, given the strong connection between income and comorbidity, it is important to explore the association between SES and all-cause mortality as well as cancer-specific mortality. Third, prior analyses have tended to use single-domain SES measures (income, education, poverty, etc) or create overly simplistic composite neighborhood SES measures.,,,,,,, However, neighborhood-level SES remains underexamined in general given that these approaches may not accurately capture the whole spectrum of factors that contribute to neighborhood SES. In contrast, the area deprivation index (ADI) consists of 17 measures of education, employment, housing quality, and poverty originally extracted from long-form US Census data, and offers a theoretical advantage. One study found a higher ADI (neighborhood disadvantage) to be associated with worse survival after cancer diagnosis, but these patients were diagnosed between 1988 and 1999, and investigators did not adjust for cancer stage or treatment in the analysis. Thus, new studies are urgently needed to apply more recent cancer data with comprehensive adjustment for factors influencing prognosis.

3To better assess cancer outcomes in the context of neighborhood and individual socioeconomic disadvantage, we used (1) the ADI,, a comprehensive composite measure of neighborhood SES, and (2) Medicare-Medicaid dual eligibility (DE), a measure of patient-level low income status. Although DE may not reflect the whole spectrum of individual-level SES, it captures one of the most important aspects of individual SES: personal income. In addition, it has been consistently validated as an informative and practical tool, especially given that individual-related SES information is frequently unavailable in large national cancer databases., Using the Surveillance, Epidemiology, and End Results (SEER)–Medicare linked database, we investigated the associations of ADI and DE with survival among elderly patients diagnosed with breast, prostate, lung, and colorectal cancer and assessed whether there was an interaction between individual- and neighborhood-level SES. In the United States, older patients (≥65 years) constitute 62.9% of the cancer population, and these 4 cancers are among the most commonly diagnosed cancers, accounting for 43.1% of cancer-related deaths in 2021. Our analysis may inform a better understanding of the relationship between neighborhood and individual SES and cancer outcomes and may support policies for ongoing investments in lower-resource neighborhoods and low-income households.

4The SEER-Medicare database reflects the linkage of 2 large population-based sources of data, which provide detailed demographic, clinical, and cause-of-death information about Medicare beneficiaries (aged ≥65 years) with cancer. The SEER program covers approximately 27.8% of the US population via 18 population-based cancer registries in 13 states. Medicare is a national health insurance covering individuals aged 65 years or older and those younger than 65 years with special medical needs. We used SEER-Medicare data to identify patients diagnosed with nonmetastatic breast (International Classification of Diseases for Oncology, Third Edition [ICD-O-3] code: 26000), prostate (ICD-O-3 code: 28010), lung (ICD-O-3 code: 22030), and colorectal cancer (ICD-O-3 codes: 21041-21052) between January 1, 2008, and December 31, 2011. Rather than using more recent years of diagnosis, we selected this time frame to give sufficient follow-up time, because our sample comprises patients with nonmetastatic diseases who may live for many years after cancer diagnosis. In addition, patients must have been enrolled in Medicare Parts A and B fee-for-service coverage from 2 years prior to diagnosis through 1 year after. This time frame ensured that we would have access to complete claims in order to assess prediagnosis comorbidities and receipt of cancer treatment. Patients must have had at least 1 claim billed to Medicare during this fee-for-service enrollment window. Thus, included patients were age 67 years or older at time of diagnosis. eFigure 1 in the Supplement represents the derivation of the final included populations for each cancer. The Yale Institutional Review Board approved this study as exempt human research because SEER-Medicare data were deidentified, and informed consent from patients was not required. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

5We used the latest version of the ADI available at the time of this study (2018) to measure patients’ neighborhood SES via linked ZIP codes. The ADI ranges from 1 to 100 and is presented in national percentile rankings, with higher scores indicating higher levels of neighborhood socioeconomic disadvantage. In line with previous studies,,,,, we categorized the ADI measure into quintiles within each cohort. Quintile 1 corresponded to the most affluent neighborhoods, whereas quintile 5 referred to the most deprived neighborhoods.

6Individual-level SES was determined based on DE for Medicare and Medicaid. Among Medicare beneficiaries, some patients with limited income and resources may be jointly enrolled in Medicaid if they meet eligibility criteria determined by their state of residence. Medicaid may help them cover costs and services not available through Medicare, such as long-term nursing facility services and home health services. Dual eligibility has been consistently used as an indicator for low income status., In this study, patients were considered dually eligible if they had at least 1 month of DE in the year prior to diagnosis.

7The primary outcome was overall survival, defined as time from cancer diagnosis to death from any cause or the end of the study (December 31, 2017), whichever came first. The secondary outcome was cancer-specific survival. Given that cause of death was not available for the year 2017, we set December 31, 2016, as the end of follow-up for cancer-specific survival analyses and censored patients at death if they died from other diseases rather than primary cancer.

8The characteristics of patients included age at diagnosis (67-69, 70-74, 75-79, 80-84, or 85-94 years), sex (male or female), race and ethnicity (Black, Hispanic White, non-Hispanic White, or other [including Asian, Native American, races not identified as previous groups, and unknown]), marital status (yes, no, or unknown), cancer stage (I, II, III), Elixhauser comorbidity index (0, 1-2, or ≥3), surgery (yes or no), chemotherapy (yes or no), and radiotherapy (yes or no). Additionally, hormone receptor status (estrogen and progesterone receptors) was available for breast cancer and categorized as (1) either estrogen receptor or progesterone receptor was positive, (2) both estrogen receptor and progesterone receptor were negative, and (3) unknown. For prostate cancer, androgen deprivation therapy (yes, no) was also available. Race and ethnicity were commonly collected and identified by each cancer registry in the database. In this cohort, the overwhelming majority of Black patients were non-Hispanic; thus, non-Hispanic and Hispanic Black patients were combined as Black.

9Patients’ characteristics were compared by ADI quintiles using the χ2 test for categorical variables and the Kruskal-Wallis test for continuous variables. Among each cancer type, differences of overall and cancer-specific survival between quintiles were assessed via the Kaplan-Meier method and the log-rank test., Only Q1, Q3, and Q5 are presented in Table 1 owing to space limitations.

10We applied Cox proportional hazards regression to estimate hazard ratios (HRs) for the associations of ADI and DE with overall mortality and cancer-specific mortality. Regarding biological plausibility and change-in-estimate criterion (10% cutoff),, the following were included as potential covariates: age, sex, marital status, cancer stage, Elixhauser comorbidity index, surgery, radiotherapy, and chemotherapy. Furthermore, hormone receptor status was considered for breast cancer, and androgen deprivation therapy was included for prostate cancer. Missing data for categorical variables were replaced with the level of “unknown,” whereas no missing data were observed for continuous variables. To calculate adjusted HRs, ADI and DE were both included in the models in addition to potential covariates. For sensitivity analysis, we also calculated adjusted HRs of ADI and DE without mutual adjustment. We used the Schoenfeld residuals method to test proportional hazards assumption and did not detect violations.

11To explore potential interactions between ADI and DE, we considered ADI as an ordinal variable, and the level of ADI quintiles (1 through 5) corresponded to the equivalent integer value (1 through 5). By DE status (yes vs no), we further estimated adjusted HRs of ADI by the increase of 1 quintile among 2 subgroups (DE vs non-DE beneficiaries) and assessed the interaction term using likelihood-ratio tests. In addition, we calculated distribution of DE by ADI and included this 2-way tabulation as eTable 1 in the Supplement.

12We conducted all statistical analyses using SAS statistical software, version 9.4 (SAS Institute Inc) and R, version 4.0.2 (R Foundation for Statistical Computing) from January 23 to April 15, 2021. All P values were 2-sided, and P <.05 was considered as statistically significant. Point estimates were presented with 95% CIs. Multiple comparisons were not adjusted.

13A total of 96 978 patients were analyzed. Of 25 968 female patients with breast cancer, median age at diagnosis was 76 years (IQR, 71-81 years); 1603 (6.2%) were Black, 21 725 (83.7%) were non-Hispanic White, and 2640 (10.2%) were of other race and ethnicity, including Hispanic; and 9545 (36.8%) died (median follow-up, 7.1 years [IQR, 5.5-8.4 years]). Of 35 150 male patients with prostate cancer, median age at diagnosis was 73 years (IQR, 70-77 years); 2884 (8.2%) were Black, 27 001 (76.8%) were non-Hispanic White, and 5265 (15.0%) were of other race and ethnicity, including Hispanic; and 10 431 (29.7%) died (median follow-up, 7.4 years [IQR, 6.3-8.7 years]). Of 16 684 patients with lung cancer, median age at diagnosis was 76 years (IQR, 71-81 years); 8412 (50.4%) were women; 1162 (7.0%) were Black, 13 926 (83.5%) were White, and 1596 (9.6%) were of other race and ethnicity, including Hispanic; and 13 984 (83.8%) died (median follow-up, 1.6 years [IQR, 0.5-5.1 years]). Of 19 176 patients with colorectal cancer, median age at diagnosis was 78 years (IQR, 72-84 years); 10 486 (54.7%) were women; 1233 (6.4%) were Black, 15 557 (81.1%) were White, and 2386 (12.4%) were of other race and ethnicity, including Hispanic; and 11 378 (59.3%) died (median follow-up, 6.1 years [IQR, 2.2-7.8 years]).

14Regardless of cancer types, compared with those in the most affluent neighborhoods, patients living in the most disadvantaged neighborhoods were more likely to be Black, to be unmarried, to be DE beneficiaries, to have more comorbidities, and to die from any cause or primary cancer by the end of follow-up (Table 1). Patients with breast, lung, and colorectal cancer in the most disadvantaged neighborhoods tended to be younger and diagnosed at stage III (although stage was not significant for colorectal cancer). In contrast, among those with prostate cancer, those patients in the most disadvantaged neighborhoods were more likely to be older and have an early cancer stage. No clear patterns of differences in utilizing surgery, chemotherapy, and radiotherapy were observed among these cancer types across neighborhoods (Table 1).

15Compared with those in the most affluent neighborhoods, patients living in the most disadvantaged neighborhoods had worse overall survival (Figure 1). For 5-year overall survival (Table 2), the estimate difference ranged from 8.1% (colorectal cancer) to 11.2% (prostate cancer). We observed a similar pattern for cancer-specific survival (Table 2; eFigure 2 in the Supplement), but the magnitudes of survival differences were noticeably different across cancer types. For example, the absolute percentage point difference in cancer-specific survival between quintile 1 and quintile 5 was small for prostate cancer (2.0%) but quite large for lung cancer (10.1%).

16Among all cancer types, living in the most disadvantaged neighborhoods was significantly associated with unadjusted higher risk of overall mortality compared with living in the most affluent neighborhoods (eTable 2 in the Supplement). After comprehensive multivariable adjustment (Table 3), the associations were attenuated but remained statistically significant (adjusted HRs: 1.34 [95% CI, 1.26-1.43] for breast, 1.51 [95% CI, 1.42-1.62] for prostate, 1.21 [95% CI, 1.14-1.28] for lung, and 1.24 [95% CI, 1.17-1.32] for colorectal cancer). We observed a similar pattern for DE status, which was significantly associated with higher risk of overall mortality (adjusted HRs: 1.22 [95% CI, 1.15-1.29] for breast, 1.29 [95% CI, 1.21-1.38] for prostate, 1.14 [95% CI, 1.09-1.20] for lung, and 1.23 [95% CI, 1.17-1.29] for colorectal cancer).

17For cancer-specific survival, we detected a similar pattern. Across the 4 cancer types, living in the most disadvantaged neighborhoods and being a DE beneficiary were significantly associated with higher risk of dying from primary cancer (Table 2; eTable 2 in the Supplement).

18In sensitivity analyses, regardless of cancer types, simultaneously adjusting for both ADI and DE had little impact on the independent effect of each measure on survival (Table 3, eTable 3 in the Supplement). Additionally, of the 8 interactions of ADI and DE for overall and cancer-specific mortality among 4 cancer types (Figure 2), only 1 was significant, and the difference had limited clinical significance (HR, 1.08; 95% CI, 1.05-1.11 vs HR, 1.04; 95% CI, 1.03-1.06) among DE vs non-DE colorectal cancer patients, for adjusted HRs of ADI with overall mortality. Thus, these analyses may suggest that ADI and DE were independently associated with overall survival and cancer-specific survival.

19In this study, we found that neighborhood and individual socioeconomic disadvantage were significantly and independently associated with worse overall and cancer-specific survival among patients with nonmetastatic breast, prostate, lung, and colorectal cancer.

20Our study complements and expands on prior work in many important ways. First, instead of applying single-metric or one-off composite neighborhood SES features,,,,,,,, we used the most recent version of ADI, a comprehensive composite measure to assess the complexity of neighborhood SES. Second, although individual SES information is frequently unavailable in large national databases,, DE status among Medicare beneficiaries has been uniquely identified as a specific measure of patient-level low income,, and we relied on it to reflect individual socioeconomic disadvantage. Third, using recent real-world big data, we examined multiple common cancers and assessed how neighborhood and individual socioeconomic disadvantage may together result in worse cancer outcomes, even after accounting for demographic, clinical, pathological, and treatment factors. The magnitudes of neighborhood and individual socioeconomically disadvantaged status are comparable to the effects of contemporary adjuvant therapies for these common cancers.,,, Thus, along with investing in treatment development, these novel findings support the need for policies for ongoing investments in disadvantaged neighborhoods and low-income households.

21This study is among the first to our knowledge to use national data, as most prior studies investigated the link between ADI and cancer using primarily regional databases. Neighborhood disadvantage has been reported to be associated with (1) increased cancer prevalence, incidence, and mortality,; (2) obesity and inferior overall survival among pediatric patients with acute lymphoblastic leukemia,; (3) lack of adjuvant therapy among patients with localized pancreatic cancer; and (4) anxiety among patients with advanced cancer. One SEER study reported higher ADI (neighborhood disadvantage) to be associated with worse overall survival among cancer patients diagnosed between 1988 and 1999: HRs (highest decile vs lowest decile) were 1.68 (95% CI, 1.57-1.79) for breast, 1.57 (95% CI, 1.46-1.68) for prostate, 1.29 (95% CI, 1.23-1.36) for colorectal, and 1.56 (95% CI, 1.54-1.59) for all cancers. However, such survival estimates were not adjusted for cancer stage and treatment and may not accurately reflect the associations of neighborhood disadvantage with cancer survival, given that stage and treatment are key factors influencing the prognosis of cancer patients. From 55 clinical trials for 15 cancer types (1985-2012), 1 recent study identified 41 109 patients with cancer and combined them as a single cohort for the analysis. Compared with the lowest ADI quintile, patients with cancer living in the most deprived neighborhoods (highest quintile) had worse overall survival (HR, 1.28; 95% CI, 1.20-1.37), progression-free survival (HR, 1.20; 95% CI, 1.13-1.28), and cancer-specific survival (HR, 1.27; 95% CI, 1.18-1.37), suggesting that even in this group of trial participants who initially all had access to high-quality cancer care, there were still survival disparities across strata of neighborhood SES. Given that all cancer types were combined and separate analyses were not conducted, it was unclear which cancers were primarily subject to neighborhood socioeconomic disadvantage. In addition, only 8% of American adult cancer patients enrolled in clinical trials owing to multiple structural and clinical barriers for trial participants., Thus, the generalizability of such results has not been examined until now. Using a large national cancer database, our findings may be generalizable to a broader neighborhood and cancer population across the country. Our study adds to prior findings by demonstrating that (1) survival disparities persist after adjustment for demographic, clinical, pathological, treatment, and individual SES factors in real-world data; and (2) the survival of patients with the 4 most common cancers is subject to neighborhood disadvantage as measured by the ADI.

22Furthermore, in comparing 5-year cancer-specific survival across the most affluent and deprived neighborhoods, we noticed a small difference for prostate cancer but a large one for lung cancer. In this study, at least 90% of prostate cancer cases were diagnosed at stage I and cancer-specific survival was at least 96% regardless of neighborhood SES, suggesting that patients were very likely to be cured from primary prostate cancer. However, patients with lung cancer living in the most deprived neighborhood were more likely to be diagnosed at stage III and less likely to get surgery, which was considered to be the most curative treatment. Thus, a large difference in cancer-specific survival between neighborhoods was observed among patients with lung cancer. Future studies in other cancer types are also needed to explore the differences of cancer-specific survival estimates between neighborhoods.

23In large national cancer databases,, individual-level SES information is frequently unavailable, whereas such information is available in the SEER-Medicare database. Although DE has been consistently considered as a reliable indicator for low income,, only 4 prior studies have looked into DE status and cancer survival: 3 used state-level data (Ohio, Michigan, and North Carolina), and only 1 included all SEER states.,,, In these studies, DE status was reported to be associated with inferior overall survival after diagnosis of prostate, lung, and gynecologic cancer,,, and worse colorectal cancer–specific survival. Owing to limited prior work and lack of replicability, our study complements previous studies by evaluating the most common cancers through recent national SEER-Medicare data and reports the associations of individual socioeconomic disadvantage with inferior survival among patients with breast, prostate, lung, and colorectal cancer. Thus, in addition to adjusting for neighborhood-level SES, future studies using national cancer databases should also adjust for individual-level SES measures if they are available or could be obtained from external resources.

24Cancer treatment confers a large financial burden on patients and the health system., Although patients’ medical expenses may be largely covered by Medicare and/or Medicaid, we still observed persistent survival disparities associated with neighborhood and individual SES disparities, even after accounting for cancer treatment. The recent COVID-19 pandemic has exacerbated long-standing economic and health inequalities, and has exposed cancer patients, especially those from neighborhood and individual socioeconomically disadvantaged backgrounds, to greater risk of death., Health inequities in cancer survival related to neighborhood and individual SES status are symptoms of deeply rooted structural and systematic barriers in policy, education, health care, employment, insurance, and the justice system, as well as underlying racism and classism. For patients living in poverty and/or in deprived neighborhoods, these barriers can manifest in many ways that negatively influence cancer survival,,,,,,,, including financial toxicity (ie, difficulty related to the cost of medical care) after cancer diagnosis, lack of access to physicians and health care resources, unequal treatment in health care, inefficacious referral systems, mistrust between providers and patients, poor social support network, and barriers to travel for initial and follow-up care. Our findings may support policies for ongoing investments in disadvantaged neighborhoods and low-income households, which could increase their opportunities for healthy and safe living and expand their access to timely cancer treatment.

25This study has several limitations. Although ADI is a comprehensive composite measure to assess neighborhood socioeconomic disadvantage, it may not reflect every aspect of neighborhood SES. However, in line with other studies using alternative composite measures of neighborhood SES (such as the Yost index), inferior survival was observed among patients with breast, prostate, lung, and colorectal cancer.,,, A similar limitation also applies to DE status. Low income is an important attribute of disadvantaged SES, but it may not manifest each facet within individual SES. However, our findings add to limited prior work and replicability in research. Additionally, we focused only on the 4 most commonly diagnosed cancers; thus, our findings may not be generalizable to other cancer types. Furthermore, there are some important limitations to the SEER-Medicare database. Our analysis was restricted to older Medicare beneficiaries with fee-for-service coverage, so our results may not be generalizable to younger patients or those with other types of insurance.

26The findings of this cohort study suggest that patients with breast, prostate, lung, and colorectal cancer experienced worse survival if coming from neighborhood or individual socioeconomically disadvantaged backgrounds. Such disparities persisted even after accounting for demographic, clinical, pathological, and treatment factors. Policies for ongoing investments in low-resource neighborhoods and low-income households are needed to improve cancer outcomes and reduce health disparities. Future studies should examine other common cancers to help health professionals and policy makers better understand the differences in cancer outcomes.