Effects of socioeconomic status on esophageal adenocarcinoma stage at diagnosis, receipt of treatment, and survival: A population-based cohort study

The incidence of esophageal adenocarcinoma (EAC) is increasing worldwide and has overtaken squamous histology in occurrence. We studied the impact of socioeconomic status (SES) on EAC stage at diagnosis, receipt of treatment, and survival. A population-based retrospective cohort study was conducted using Ontario Cancer Registry-linked administrative health data. Multinomial logistic regression was used to examine the association between SES (income quintile) and stage at EAC diagnosis and EAC treatment. Survival times following EAC diagnosis were estimated using Kaplan-Meier method. Cox proportional-hazards regression analysis was used to examine the association between SES and EAC survival. Between 2003–2012, 2,125 EAC cases were diagnosed. Median survival for the lowest-SES group was 10.9 months compared to 11.6 months for the highest-SES group; the 5-year survival was 9.8% vs. 15.0%. Compared to individuals in the highest-SES group, individuals in the lowest-SES category experienced no significant difference in EAC treatment (91.6% vs. 93.3%, P = 0.314) and deaths (78.9% vs. 75.6%, P = 0.727). After controlling for covariates, no significant associations were found between SES and cancer stage at diagnosis and EAC treatment. Additionally, after controlling for age, gender, urban/rural residence, birth country, health region, aggregated diagnosis groups, cancer stage, treatment, and year of diagnosis, no significant association was found between SES and EAC survival. Moreover, increased mortality risk was observed among those with older age (P = 0.001), advanced-stage of EAC at diagnosis (P < 0.001), and those receiving chemotherapy alone, radiotherapy alone, or surgery plus chemotherapy (P < 0.001). Adjusted proportional-hazards model findings suggest that there is no association between SES and EAC survival. While the unadjusted model suggests reduced survival among individuals in lower income quintiles, this is no longer significant after adjusting for any covariate. Additionally, there is an apparent association between SES and survival when considering only those individuals diagnosed with stage 0-III EAC. These analyses suggest that the observed direct relationship between SES and survival is explained by patient-level factors including receipt of treatment, something that is potentially modifiable.

Introduction of EAC diagnosis was the same as the date of death or individuals whose EAC was not the primary site were excluded.

Data sources
For this study, we used the following databases: the OCR, the Ontario Health Insurance Plan (OHIP), the Registered Persons Database, and the Canadian Institute for Health Information Discharge Abstract Database (DAD) and the National Ambulatory Care Reporting System (NACRS). The OCR contains all cancer incidence (new cases) and mortality (deaths) in Ontario from 1964 onward. OHIP data contains the records of all physician billings for inpatient and outpatient visits and procedures starting from January 1991. The Registered Persons Database contains demographic and address information for all people registered for provincial government-sponsored health insurance coverage. The DAD contains demographic, clinical, and administrative data for hospital admissions and day surgeries in Ontario starting from 1991, and the NACRS contains administrative, demographic, clinical, and financial data for hospital-based ambulatory care. To track treatments use, we searched for claims for surgical resection, chemotherapy and radiotherapy for EAC from OHIP, DAD and NACRS fee codes. We also searched for fee codes from OHIP, DAD and NACRS for the following palliative procedures: esophageal dilation, drainage, esophageal stenting, laser debulking of tumor, and palliative care. See Supporting Information S2 Table. Data were provided by the Institute for Clinical Evaluative Sciences which contain the health records for the roughly 14 million residents of Ontario. All data provided on EAC cases were post cancer diagnosis. No health records prior to cancer diagnosis were provided. We therefore could not assess screening for EAC (endoscopy and biopsy), prior Barrett's esophagus diagnosis, or prior cancer diagnosis.
Individual-level SES was not available therefore area-level SES was used as a surrogate. Area-level SES was quantified using median neighbourhood household income. Median neighbourhood household income was determined through linking of postal codes to Canadian census data; income was categorized into quintiles corresponding to income status of neighbourhoods. Income quintile 1 represents the lowest 20% of neighborhoods and income quintile 5 represents the most well-off 20% of neighbourhoods.
Ontario is currently divided into 14 health regions which plan and fund local health care. Each Local Health Integration Network's mandate is to make the health system more efficient and improve access to quality care [25]. Patients' local health regions were used as a factor to explain regional health care service and availability.
We classified patient comorbidity using the Johns Hopkins Adjusted Clinical Groups (ACG) case-mix system, which has been validated in the United States [26,27] and in Canada [28,29]. The ACG system measures individuals' morbidity by grouping individuals based on their age and gender and all medical diagnoses over a given time period. For this study, we used Ontario inpatient Discharge Abstract Database and outpatient OHIP diagnosis codes from the year prior to the date of EAC diagnosis to estimate case-mix. Patients grouped into 32 different ADG categories method may be useful for comorbidity adjustment in administrative health care data when comparing morbidity, mortality, or health care utilization and costs [30][31][32].

Outcome measure
The primary outcome for our study was survival time after EAC diagnosis. Survival time was calculated using the time between death and date of diagnosis. If no death was observed during the follow up period, the patient was censored. As secondary outcomes, stage of diagnosis, receipt of EAC treatment and health region were compared by SES.

Statistical analysis
Overall EAC patient characteristics and patient characteristics by SES (income quintile) and by year of EAC diagnosis from 1993 to 2012 were tabulated. Chi-squared tests were used to examine the association between income quintile and relevant variables as mentioned above. The following survival analyses were determined for study periods 2003-2012 due to the availability of EAC stage. Median survival times (months, with interquartile range [IQR]), and 1-year, 3-year, and 5-year survival (95% confidence intervals [CIs]) after EAC diagnosis overall, and stratified by income quintile and other covariates were estimated using Kaplan-Meier survival analysis. Differences between survival times were assessed using log-rank tests.
Multinomial logistic regression analysis was used to examine the association between income quintile and stage at EAC diagnosis, yielding odds ratios and corresponding 95% CIs of stage II, stage III, or stage IV relative to stage 0-I EAC at diagnosis, using income quintile 5 as the reference. Multinomial logistic regression analysis was also used to assess the association between income quintile and receiving treatment for EAC relative to no treatment as well as between income quintile and patients' health regions relative to Central region. We adjusted for potential confounding covariates, including age, gender, urban/rural residence, birth country, health region (to assess the association between income quintile and cancer stage and treatment), ADGs, and year of EAC diagnosis. Cancer stage at diagnosis was included as a covariate to assess the association between income quintile and EAC treatment and health region.
Cox proportional-hazards regression analysis using unadjusted (univariate) and adjusted (multivariate) models were used to assess the association between income quintile and EAC survival. The main exposure variable we assessed was SES. Age and gender were evaluated as confounders of SES, along with urban/rural residence, birth country, health region, ADGs, cancer stage at diagnosis, treatment, and year of EAC diagnosis. EAC treatment was modeled as a time-dependent variable within the proportional-hazards regression model; treatment status changed from 0 to 1 based on the treatment date (if cases received treatment during illness). The results are reported as hazard ratios (HRs) with 95% CIs. Additionally, the association between income quintile and survival was assessed, considering only those individuals diagnosed with stage 0-III EAC (excluding advanced-stage IV).
A two-sided P-value of < 0.05 was considered to be statistically significant. Statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and STATA version 12.0 (Stata Corporation, College Station, TX) statistical software applications.

Ethics approval
Ethics approval for the study was granted by the University of Toronto Health Sciences Research Ethics Board. Informed consent was not obtained because this secondary analysis accessed existing de-identified data; consent was therefore deemed to be neither feasible nor necessary.

Baseline characteristics of patients diagnosed with EAC
A flow chart of the study population can be found in S1 Fig. Overall  The proportion of cases diagnosed among patients aged 60-64 years increased from 11.8% during 1993-1997 to 16.2% during 2008-2012; conversely, the proportion of cases diagnosed among those aged 70-74 years decreased (from 18.0% to 13.2%) during this same time period. The majority of patients were male (84%), with a male to female ratio of about 5:1. The highest number of EAC cases occurred in persons with ADGs 11 and above (which increased from 43.2% during 1993-1997 to 49.1% during 2008-2012) (S3 Table).

Population with EAC diagnosis by SES
Descriptive characteristics of the 5,382 individuals with EAC stratified by income quintile are summarized in Table 1. Urban/rural residence (P < 0.001), Ontario health region (P < 0.001), and receiving EAC treatment (P = 0.048) were the factors that were significant when stratified by income quintiles. People within lower income quintiles were less likely to receive treatment (no treatment: 41.5% and 39.4% for lower income quintiles 1 and 2, respectively) than those within higher income quintiles (36.3% to 34.1% for income quintiles 3-5, P = 0.002). In addition, people in lower to mid SES groups (84.1% to 85.8%) were more likely to die than those in higher income quintiles (81.7% to 82.3%, P = 0.066), although this was not statistically significant.  There was no significant difference in the overall survival between the income quintiles (log-rank test P = 0.085), Fig 1. The 5-year survival rate seemed different between the income quintiles, with income quintile 1 having a survival rate of 9.8% after 5 years post-diagnosis of EAC compared to 15.0% for income quintile 5. When the survival curves were stratified by EAC stage, there was a significant difference in the survival between the income quintiles according to stages II (P = 0.005), III (P = 0.045), and IV (P = 0.045) (Fig 2A-2D). When survival curves were stratified by treatment type, there was no significant difference in survival times for the income quintiles (Fig 3A-3G).

Association between SES and stage at EAC diagnosis, EAC treatment and health region
There was not a significant association between SES and cancer stage at EAC diagnosis, after adjusting for age, gender, residence, birth country, health region, ADG, and year of EAC diagnosis (Table 3). Table 4 shows the association between SES and EAC treatment received. Although there are some differences (Table 1), the statistical test for an overall association did not achieve significance.
There was a significant difference between SES income quintiles among Ontario health regions when compared to Central region, except for Erie St. Clair, Waterloo Wellington, Mississauga, Toronto Central, and North West (S4 Table).

Association between SES and EAC survival
Within the unadjusted Cox proportional-hazards model, patients with EAC in the lower three income quintiles had increased risk of mortality relative to the highest income category (Table 5); those in the lower income quintiles (Q1-Q2) experienced a 16%-17% increase in the risk of death (Q1: HR = 1.16; 95% CI, 1.00-1.36; Q2: HR = 1.17, 95% CI, 1.00-1.37). However, this association disappeared in the fully-adjusted multivariate model; there was no significant association between SES and EAC survival after controlling for age, gender, residence, birth country, health region, ADG, cancer stage, treatment, and year of diagnosis. Additionally, increased mortality risk was observed for age (P = 0.001), cancer stage at EAC diagnosis (P < 0.001), chemotherapy (P < 0.001), radiotherapy (P < 0.001), surgery plus chemotherapy (P < 0.001), and year of EAC diagnosis. However, including only individuals with stage 0-III cancer, and excluding those patients with more advanced-stage IV EAC, there was a significant association between SES and EAC survival after controlling for covariates (S5 Table).

Sensitivity analysis
After multiple imputation for variables with missing data such as income quintile, urban/rural residence, birth country and cancer stage at EAC diagnosis, and after adjusting for confounding covariates, there was also no significant association between SES and stage at EAC diagnosis (S6 Table) or EAC treatment (S7 Table). Additionally, in the fully-adjusted multivariate model, patients with EAC in the lower two income quintiles had a significantly increased risk of mortality relative to the highest income category (S8 Table). When including only individuals with stage 0-III cancer, and excluding those patients with more advanced-stage IV EAC, Total N = 1,573. Multinomial logistic regression analysis (fully-adjusted model) overall P-values: income quintile (P = 0.209), age (P < 0.001); gender (P = 0.665); residence (P = 0.178); birth country (P = 0.193); Ontario health region (P = 0.008); Aggregated Diagnosis Group (ADG) (P = 0.635); cancer stage at EAC diagnosis (P < 0.001); and year of EAC diagnosis (P < 0.001).
https://doi.org/10.1371/journal.pone.0186350.t004 patients with EAC in the lower four income quintiles had a significantly increased risk of mortality relative to the highest income category (S9 Table).

Discussion
This population-based retrospective cohort study examined the effects of SES on stage of diagnosis of EAC, receiving treatment, regional variation (Ontario health region), and survival.
The results indicate that individuals in lower SES categories have reduced survival compared to those in the highest income quintile, but these differences disappear after adjusting for confounders. While there is an apparent 17% increase in mortality for individuals in the lower income quintiles compared to the highest, the significance of this association disappears to almost null in the fully-adjusted regression model, after controlling for additional covariates. However, the significant association between SES and EAC survival remained when considering only those who presented with stage 0-III EAC at diagnosis. There was no significant association between SES and EAC stage at diagnosis or between income quintile and receipt of potentially curative EAC treatment.
A previous study in Ontario found no significant relationship between SES and stage at diagnosis for hepatocellular carcinoma [36]; other Canadian studies have yielded mixed results with respect to various cancers [37,38]. Conversely, studies in the United States have found a significant relationship between SES and stage at diagnosis [39]. This may be explained by  [40][41][42][43]. A study by Gorey et al. that compared the effect of SES on colon cancer and treatment in San Francisco and Toronto found that differences in SES had a more pronounced influence on survival in California compared to Ontario, with Canada's universal health care system functioning more equitably for both rich and poor [41]. Other studies including one published in 2010 by Booth et al. found that a survival gradient is in fact present between individuals of differing SES [37]. One presented explanation was the increased likelihood of higher SES individuals to receive curative treatment for hepatocellular carcinoma [19]. Corresponding to our study, a population-based study of patients with potentially resectable esophageal cancer in the Netherlands conducted by Koeter et al. found that surgical resection occurred less often among less well-off patients; however, survival was not significantly affected by SES [44].
A previous study conducted by Tinmouth et al. in 2011, found that compared to the Central region, people in the North West Ontario health region were 6.5 times as likely to contract EAC [45]. Possible explanations for this phenomenon include ethnic variation and suboptimal treatment center placement. Our study identified differences in SES quintiles by health region, however, the adjusted proportional-hazards model findings observed no association between health region and EAC survival. Differences in the health region distribution of EAC cases may also be explained by physician supply. A previous study conducted in Ontario found that access to regional primary care physicians was significantly correlated with health outcomes [46,47]. These results indicate a need to further research the health effects of physician supply and the efficacy of current health care system distribution patterns.
Our study was large and population-based. It has some limitations, however. The study design was retrospective and so cannot establish causation. SES was an ecological variable and may not be fully indicative of true individual-level SES. Median neighborhood household income also fails to account for several other important determinants of SES including social status, employment type, and social capital. We also could not account for several primary risk factors for EAC, including GERD symptoms, Barrett's esophagus, tobacco smoking, race or ethnicity, and alcohol consumption, due to a lack of data. Population distributions of these risk factors may accompany disparities in SES. European studies have indicated that lower SES is associated with worsened occurrence of GERD symptoms [48,49]. Conflicting results have been reported regarding Barrett's esophagus; a study in the United States found that greater education is associated with a reduced likelihood of Barrett's esophagus [50], while a United Kingdom study found that increased Barrett's esophagus risk accompanies increased SES [9]. In Canada, increased tobacco use is observed among those at lower SES levels [51], while United Kingdom researchers posit that alcohol use leads to more deleterious health impacts among populations experiencing greater social deprivation [52]. We could not assess screening for EAC (endoscopy and biopsy), prior Barrett's esophagus diagnosis, or prior cancer diagnosis. EAC patients with a prior Barrett's esophagus diagnosis are commonly diagnosed with earlier stage disease and have improved survival compared with EAC patients with no prior Barrett's esophagus diagnosis [53]. We were not able to discern whether patients did not receive treatment because they were not offered it, had poor functional status, declined it, or experienced other barriers [54]. Lastly, stage at diagnosis was missing for a lot of patients and they were excluded from the primary analysis. Statistically, multiple imputation is an established method to deal with replacing each missing value with a set of plausible values to ensure the results are unbiased and capture the appropriate degree of precision.

Conclusion
We saw a direct association between SES and EAC survival in Ontario, but this could be explained by patient-level confounders, including receipt of treatment. This indicates that even in our universal health care system there are inequities that affect survival. Further work needs to be done to identify the reasons for any barriers to treatment and find ways to overcome these barriers.