Socioeconomic Factors and Risk of Esophageal Adenocarcinoma: A Nationwide Swedish Case-Control Study

The incidence of esophageal adenocarcinoma has over the past three decades been increasing rapidly, whereas the incidence of gastric cardia adenocarcinoma has increased moderately, in the United States, several European countries, Australia, and New Zealand (1-3). The reasons for this striking trend remain unknown (2, 4). Gastroesophageal reflux (5, 6), high body mass index (BMI; refs. 7, 8), and male gender (male/female ratio, 7:1; refs. 3, 9) are the most established risk factors, but the rising incidence cannot be explained by changes in the prevalence of these risk factors only (10). The suddenness and rapidity of the rising incidence of esophageal adenocarcinoma (11) suggests that environmental exposures, introduced before the 1970s, are of importance in explaining the increasing incidence (12, 13). Furthermore, the descriptive epidemiology of esophageal cancer suggests that, unlike esophageal squamous cell carcinoma (SCC), esophageal adenocarcinoma may be a disease mainly affecting the well-off. This is indicated by the continued predominance of esophageal SCC in developing countries (3, 14), by the increases of esophageal adenocarcinoma seemingly confined to the increasingly affluent Western world, and by the steeper increase among U.S. Whites than non-Whites (1, 4). In contrast, gastric Helicobacter pylori infection, linked to poor living conditions during childhood (15), has repeatedly been shown to reduce the risk of esophageal adenocarcinoma (16-18), albeit not confirmed by all studies (19).

Associations between socioeconomic status (SES) and morbidity, such as cancer, have been observed in several studies (20-22). Income, education, and occupation are widely accepted measures of social class (22). Although low SES has been repeatedly linked to an increased risk of esophageal SCC (23, 24), independent of the dominating risk factors for this cancer in Western populations (i.e., tobacco and alcohol; ref. 9), few studies have investigated socioeconomic factors in relation to esophageal or cardia adenocarcinoma separately. There are some indications that low SES, based on measures of income and education, might also increase the risk of these tumors (25-28), but inconsistent results (27, 29) and the use of different measures of SES make comparisons and interpretations difficult.

The aim of our study was to explore and clarify possible associations between socioeconomic factors, particularly focusing on SES based on occupation, and the separate risks of esophageal and cardia adenocarcinoma and esophageal SCC and to study if possible associations could be explained by known risk factors, including H. pylori seronegativity.

The design of our nationwide Swedish population-based case-control study has been described previously in detail (6, 18). The study encompassed the whole native Swedish population ages <80 years and living in Sweden from 1995 to 1997. All newly diagnosed patients with esophageal or gastric cardia adenocarcinoma and half of the patients with esophageal SCC (born on even-numbered dates) were eligible as cases. A complete nationwide organization for case ascertainment, including contact persons at all relevant hospital departments and collaboration with all regional tumor registries, ensured rapid identification of every potential case patient throughout Sweden. To reduce tumor misclassification, uniform routines for prospective documentation of the tumors were developed especially for the study. The diagnosis was based on a summary of the findings at endoscopy, surgery, and histopathology. One pathologist reviewed biopsies and surgical specimens or both from 97% of the patients. Cardia adenocarcinoma had its center within 2 cm oral or 3 cm aboral to the gastroesophageal junction, whereas a site proximal to 2 cm above the junction was defined as esophageal adenocarcinoma. If Barrett's esophagus was detected near the tumor, it was classified as esophageal irrespective of location, however. SCCs were classified as esophageal even if the location was the cardia.

Control persons were selected randomly from the continuously updated and complete Swedish total population register and frequency matched according to the age and sex distribution of the cases of esophageal adenocarcinoma. All study participants underwent computer-aided personal interviews by trained interviewers who were instructed to treat case patients and control persons in a strictly equal manner. The participants were asked to donate a venous blood sample. Serum was collected from a subset including 97 of 189 interviewed cases of esophageal adenocarcinoma, 133 of 262 cases of cardia adenocarcinoma, 85 of 167 cases of esophageal SCC, and 499 of 820 interviewed control persons. After centrifugation, serum samples were stored at −70°C. Antibodies to cytotoxin-associated gene A (CagA) of H. pylori infection were detected with an immunoblot assay (Helicoblot 2:1, Genelabs Diagnostics, Singapore). The laboratory analyses were done blinded to case/control status.

Informed consent was obtained from each study participant, and all regional ethics committees in Sweden approved the study.

Detailed information about lifetime occupational history was collected, including questions about employment length, workplace, and work tasks for each occupation held by the study participant for at least 1 year. Each study participant's occupational history, ranging from 1 to 10 occupations, was classified by one reviewer (C.J.) blinded for case/control status. Occupations held before 1955 were disregarded, because these were considered to be outside the relevant etiologic time window. Occupational classification was done according to the Nordic Standard Occupational Classification (NYK; ref. 30), where each occupation was coded according to a detailed five-digit classification. SES was derived from these five-digit occupational codes according to the Swedish socioeconomic classification scheme (SEI; refs. 30-32). We allocated each study participant into one of the following six well-established socioeconomic classes: (a) unskilled and semiskilled manual workers; (b) skilled manual workers; (c) assistant nonmanual employees; (d) intermediate nonmanual employees; (e) employed or self-employed professionals, higher civil servants, and executives (reference group); and (f) self-employed (other than professionals) and farmers. In the analyses, we used each study person's SES of longest duration after 1955, obtained from the occupational history.

Educational level was classified in accordance with the Swedish school system into three categories based on total number of years of education: low (0-9 years), medium (10-12 years), and high (≥13 years; reference group).

The study questionnaire included other socioeconomic dimensions: place and region of residence, duration of living with a partner, and number of children in the household. Place of residence was assessed both during childhood and during the 1970s (i.e., ∼20 years before interview) and categorized into rural (reference group) and urban or densely populated area. Region of residence during childhood and during the 1970s, respectively, was categorized into three well-defined Swedish regions: Götaland (southern Sweden), Svealand (central Sweden; reference group), and Norrland (northern Sweden). Duration of living with a partner (married/cohabitant) was categorized into <1, 1 to 10, 11 to 30, and ≥31 (reference group) years. Number of children (including oneself) living in the household during childhood was grouped into 1 to 2 (reference group), 3 to 4, 5 to 6, and ≥7 children.

To estimate relative risks, we used odds ratios (OR) and 95% confidence intervals (95% CI) estimated by conditional logistic regression (33) using the PHREG procedure in SAS (34). The regression models were conditional on the matching variables age and sex. Because socioeconomic factors are considered as “upstream” in causal pathway models of health (20), with lifestyle and biologically active factors located between the socioeconomic factor and the disease outcome, we wished to explore how the effects of the socioeconomic factors changed after adjusting for potential “downstream” factors. Thus, we wished to study to what extent downstream factors could explain any socioeconomic effects. We started with crude models evaluating each socioeconomic factor separately as a risk factor for cancer outcome. With crude estimates, we mean an estimate from a conditional logistic model conditioning on age and sex but not adjusted for any other variables. Next, we fitted multivariable models where adjustments were made for a priori known independent risk factors for the different cancer types. In the analyses of esophageal and cardia adenocarcinoma, adjustments were made simultaneously for reflux symptoms (yes/no), BMI (kg/m², in four categories based on quartiles among the controls), and tobacco smoking status (in three categories: never, previous, and current). In the analyses of esophageal SCC, adjustments were made for tobacco smoking status (in three categories: never, previous, and current) and alcohol use (in four categories: 0, 1-15, 16-70, and >70 g/wk). The influence of dietary intake of fruit and vegetables (in three categories) was thoroughly evaluated for each of the three studied cancer types; however, because no effects on the estimates of the socioeconomic factors were found, we did not include this variable in the final models. Furthermore, if we found a crude effect of a socioeconomic factor that was attenuated in the multivariable models, we fitted models including one potential mediating factor at the time to see which factor explained the most of the attenuated effect. Each exposure was evaluated using the Wald test, which considers all categories of the variable and not just pairwise comparisons to the reference category. Missing data on included variables were few (1-4%) and similar for case and control participants (Table 1). Participants with missing data on any covariate included in the models were excluded from the analyses. Furthermore, we evaluated the influence of H. pylori/CagA seropositivity (yes/no) on the estimates of the socioeconomic factors in the subset of participants who provided sera. The subset was thoroughly investigated for biases. We had data on H. pylori/CagA seropositivity for 51% of the three case groups and for 61% of the controls. Compared with the full data set, the subset was oversampled on controls, older ages, and never smokers. However, the distribution of socioeconomic factors was similar. These comparisons were done for all three cancer outcomes separately. Applying the fully adjusted model on the subset showed similar effect estimates for the socioeconomic factors compared with the full data set. Because age was controlled for by conditioning and smoking by adjustment in the models, a complete case analysis of the subset would be adjusted for any biases due to the skewed sampling (i.e., under the assumption of missing at random within age and smoking strata). Thus, for the subset H. pylori/CagA status was included as an additional covariate in the adjusted models and its influence on the estimates of the socioeconomic factors could be evaluated.

The 189 esophageal adenocarcinoma patients, 262 cardia adenocarcinoma patients, 167 esophageal SCC patients, and 820 control persons constituted 88%, 84%, 73%, and 73% of all eligible participants in the study base, respectively. Table 1 shows the reasons for nonparticipation, distribution of the studied socioeconomic factors, and other characteristics among the study participants. Unwillingness was a rare reason for nonparticipation among case patients, although it was more common among control persons. The majority of the participants were men between ages 60 and 79 years. Skilled and unskilled manual workers were the most common SES groups. There were more professionals among cardia cancer patients and controls than among patients with esophageal adenocarcinoma or SCC. The majority within all groups of participants had a low educational level, particularly evident among case patients. Most participants grew up in a rural area, although a shift was seen toward living in an urban area during the 1970s. Fewer participants lived in northern Sweden during childhood and during the 1970s compared with central or southern Sweden. The majority of all groups of participants had been married or cohabitant for three decades or more. Having lived with a partner for <1 year was more common among patients with esophageal adenocarcinoma or SCC compared with cardia cancer patients and controls. More than one third of all groups of participants grew up in a household with one or two children (Table 1).

In Tables 2, 3, and 4, we present the results from crude and adjusted models, including the covariates described in Materials and Methods. The results from the adjusted models based on the subset of participants with known H. pylori/CagA status are not shown in the tables.

In the crude model, the risk of esophageal adenocarcinoma strongly increased with decreasing levels of SES based on each individual's occupation (Table 2). We found ∼4-fold significantly increased risks of esophageal adenocarcinoma among unskilled manual workers (OR, 3.7; 95% CI, 1.7-7.7) and self-employed (OR, 3.7; 95% CI, 1.7-8.1) compared with professionals and >2-fold significantly increased risks among skilled manual workers (OR, 2.4; 95% CI, 1.1-5.3) and assistant nonmanual employees (OR, 2.3; 95% CI, 1.0-5.3). In the model adjusted for reflux, BMI, and smoking, the inverse association between SES and esophageal adenocarcinoma was weaker and the overall association was no longer statistically significant (Table 2). The attenuated effect was mainly explained by adjustment for BMI and reflux. The risk estimates were approximately similar in the model also adjusted for H. pylori/CagA status based on the subset who donated sera.

The crude analysis revealed a moderately increased risk of esophageal adenocarcinoma among persons having <9 years of education compared with persons with >12 years of education (OR, 1.8; 95% CI, 1.1-3.0), but this association disappeared in the adjusted model (Table 2). Further adjustment for H. pylori/CagA status did not influence the results. To evaluate possible confounding between SES and education, we included both SES and education in one model in which low SES remained a significant risk factor, whereas no association was seen for education (data not shown).

No statistically significant association was identified between living in an urban area during childhood compared with living in a rural and risk of esophageal adenocarcinoma. Although the crude model indicated a 30% decreased risk among persons living in an urban area during the 1970s (OR, 0.7; 95% CI, 0.5-1.0) compared with persons living in a rural, no association was seen in the adjusted models. We found a 50% to 60% reduced risk of esophageal adenocarcinoma among persons living in northern Sweden during childhood (OR, 0.6; 95% CI, 0.3-1.1) and during the 1970s (OR, 0.5; 95% CI, 0.3-0.9) compared with living in central Sweden in the adjusted model (Table 2). These results did not change after adjustment for H. pylori/CagA status.

In the crude and adjusted models, we found a >2-fold increased risk of esophageal adenocarcinoma among persons who never or for <1 year had been married or cohabitant compared with persons who had been living with a partner for ≥31 years (OR, 2.3; 95% CI, 1.2-4.5, adjusted model). There was a seemingly dose-response relation between duration of living with a partner and risk of this cancer (Table 2).

Although the point estimates indicated an inverse association between increasing number of children in the household during childhood and risk of esophageal adenocarcinoma, no statistically significant associations were found (Table 2), and adjustment for H. pylori/CagA status did not influence the results.

In the crude model, a tendency of an association between low SES and risk of gastric cardia adenocarcinoma was seen; however, in the adjusted model, it disappeared (Table 3).

A moderately increased risk of this tumor was found among persons having a low educational level compared with persons having a high educational level in the crude model (OR, 1.7; 95% CI, 1.1-2.6), but the association was attenuated when controlling for reflux, BMI, and smoking (Table 3).

In both crude and adjusted models, we found a significantly 30% to 40% decreased risk of cardia adenocarcinoma among persons living in an urban area during childhood compared with persons living in a rural (OR, 0.6; 95% CI, 0.5-0.9, adjusted model). No associations between place of residence during the 1970s or region of residence (southern, central, or northern Sweden) and risk of this cancer was found (Table 3).

We found no associations between duration of living with a partner or number of children in the household during childhood and risk of cardia adenocarcinoma (Table 3).

Adjustment for H. pylori/CagA status did not influence the results regarding the socioeconomic factors and risk of cardia adenocarcinoma.

Occupational SES was inversely associated with esophageal SCC in the crude model and the model adjusted for smoking and alcohol use. Compared with professionals, we found an ∼2-fold increased risk among all SES groups, except among skilled manual workers where an ∼4-fold significantly increased risk was found (OR, 3.9; 95% CI, 1.7-8.9), adjusted model; Table 4). In the model also adjusted for H. pylori/CagA status, the risk estimates were approximately similar.

In both crude and adjusted models, we found a 2-fold significantly increased risk of esophageal SCC among persons having a low educational level compared with persons having a high educational level (OR, 2.0; 95% CI, 1.1-3.6), adjusted model; Table 4). In a model including both SES and education, the association between SES and esophageal SCC was the strongest (data not shown). The results were not influenced by adjustment for H. pylori/CagA status.

There was a 40% decreased risk of esophageal SCC among persons living in an urban area during childhood compared with those living in a rural in the adjusted model (OR, 0.6; 95% CI, 0.4-0.9), whereas no such effect was seen regarding place of residence during the 1970s. Although the point estimates were decreased regarding living in the northern part of Sweden during childhood and the 1970s compared with living in the central part, no statistically significant associations were seen (Table 4).

In both crude and adjusted models, we found a 2- to 3-fold significantly increased risk of esophageal SCC among persons who had been living with a partner for <1 year (OR, 2.2; 95% CI, 1.1-4.4, adjusted model) or <11 years (OR, 2.7; 95% CI, 1.2-6-3), adjusted model) compared with persons who had been married or cohabitant for ≥31 years (Table 4).

No association was seen between number of children in the household during childhood and risk of esophageal SCC (Table 4), and adjustment for H. pylori/CagA status did not influence the results.

This study indicates that persons of low occupational SES and persons living without a partner for long periods of time are at a strongly increased risk of developing esophageal adenocarcinoma. A similar pattern regarding SES and duration of living with a partner was seen for esophageal SCC, whereas no such associations were found for gastric cardia adenocarcinoma.

Strengths of our study include the efforts to separate the three cancer types based on a uniform and thorough tumor classification system, the prospective and comprehensive case ascertainment that ensured identification of all new case patients occurring throughout Sweden, the population-based design and well-defined study base permitting strict random sampling of control persons, and the personal interviews done with all study participants. Moreover, our measure of SES was based on a detailed classification of each individual's occupational history. Another strength is the ability to adjust for all established risk factors, including H. pylori infection. We did not have data on use of nonsteroidal anti-inflammatory drugs, medications that might decrease the risk of esophageal cancer (35). However, in a recent study, we found that confounding by indication might explain the association between nonsteroidal anti-inflammatory drugs and esophageal cancer (36). Weaknesses are possible differential willingness to participate across SES categories and possible differential recall among case patients and control persons. The high participation rates should act against such bias, however. Moreover, the participants were not aware of our hypotheses regarding socioeconomic factors. Finally, the lack of clear associations between the socioeconomic factors and cardia adenocarcinoma provides evidence against recall bias.

The finding of an inverse association between SES and esophageal cancer warrants attention considering the widening gaps between socioeconomic groups in Sweden (37) and other countries (22). The association with esophageal adenocarcinoma are consistent with U.S. studies based on education and income as measures of SES (25-27). However, these studies are not entirely comparable for three reasons. (a) There is a slight difference in the classification of esophageal adenocarcinoma between our study and the U.S. studies, (b) esophageal and cardia adenocarcinoma were analyzed as one combined cancer type in one of these studies (25), and (c) none of the previous studies were able to adjust for all established risk factors, including reflux and H. pylori infection. Our results indicate a stronger relation with SES based on occupation compared with education. SES based on occupation could be a more informative measure than education, at least in Sweden, because persons may find employment in high positions without having a high education and because SES derived from occupation usually lasts longer and comes after the years in school, making the time window for cancer development more relevant. Previous studies have shown that occupational social class is more strongly associated with mortality compared with educational level (38). The association between occupational SES and esophageal adenocarcinoma was only partly explained by reflux, body mass, and smoking and not explained by H. pylori infection or dietary intake of fruit or vegetables. Therefore, there are probably other factors more common in low SES groups that might explain this association, speculatively including occupational exposures or infections. It has also been suggested that SES itself, regardless of associated material and economic advantages, may yield health benefits possibly through psychosocial mechanisms (39). In contrast, the finding of moderately increased risks of esophageal and cardia adenocarcinoma among persons with low education compared with persons with high seemed to be importantly attenuated by the established risk factors. This finding and our evaluation of SES and education in separate models indicates that education might be a less relevant measure of SES for these cancer types in Sweden. Furthermore, the association between low SES and an increased risk of cardia adenocarcinoma seen in previous studies (25-28) could be partly explained by insufficient adjustment for reflux, body mass, or other unmeasured factors.

An inverse association between SES and esophageal SCC has been seen in many studies (23, 24, 27). Our study suggests that low SES may act independently of the main risk factors (i.e., tobacco and alcohol use). Therefore, the underlying or mediating exposures or mechanisms are yet to be identified. However, because we analyzed smoking as a categorical exposure, residual confounding cannot be ruled out.

Our finding of elevated risks of esophageal adenocarcinoma and SCC among persons who never, or for shorter periods only, had been married or cohabitant compared with persons who had been living with a partner for decades has only been seen for esophageal SCC previously (24, 40). Studies have reported that married adults are healthier and live longer than unmarried adults (41, 42), an effect that is more apparent for men than for women (42-44). This finding is particularly interesting considering the unexplained male predominance among patients with esophageal adenocarcinoma. It has been hypothesized that marriage increases social support and income and reduces risky behavior and stress, all factors contributing to a better health (41). Our findings were not explained by the adjustment for the established risk factors, indicating potential influence of other mechanisms explaining these associations (e.g., social support). Considering dramatic increases of the number of divorces and single households in Sweden in the 20th century, our findings deserve further attention.

The finding of an inverse association between living in an urban area during childhood compared with a rural and risk of cardia adenocarcinoma could have many explanations (e.g., a higher prevalence of households with high SES in urban areas or that there are unmeasured harmful environmental exposures in the countryside not seen in densely populated areas). However, no association was seen regarding place of residence in the 1970s. No previous studies have addressed this topic before, stressing a need for cautious interpretation and more detailed measures to clarify possible associations.

In the adjusted model, a reduced risk of esophageal adenocarcinoma was found among persons living in the northern part of Sweden during the 1970s compared with persons living in the central part, and the point estimates were decreased regarding region of residence during childhood as well. This might be explained by a higher prevalence of H. pylori infection seen in these parts of Sweden (45). However, the adjustment for H. pylori/CagA status did not influence the results regarding region of residence.

From a causal perspective, the crude ORs estimate the unbiased effect of SES on esophageal adenocarcinoma in a causal model where we assume that reflux, BMI, and smoking are intermediates in the causal pathway between SES and cancer. In a causal model where we assume that reflux, BMI, and smoking precede SES in the causal pathway, these factors are confounders and the adjusted ORs provide an unbiased estimate of the relative risk of SES on cancer. However, if we do not assume any causality, our estimated associations indicate that low SES might be an independent risk factor for esophageal adenocarcinoma, acting separately from established risk factors. Interestingly, our results are similar as in studies from the United States despite large differences in health care and welfare systems between these countries. Even in a country like Sweden with equal access to health care, low SES is still strongly related to an increased risk of esophageal adenocarcinoma. A common pattern could be social hierarchies where it is relative social position or social networks within countries that matters most (46). It is of importance to find the yet unidentified mechanisms that can explain this strong association and to study if there are exposures existing predominantly within low SES groups that could explain the increasing incidence of esophageal adenocarcinoma.

In conclusion, our Swedish population-based nationwide study reveals strong influence of socioeconomic factors in the etiology of esophageal adenocarcinoma. Low occupational SES and life without a partner may strongly increase the risk of both adenocarcinoma and esophageal SCC, and these effects are only partly explained by established risk factors. These findings stress the need to investigate possible preventable underlying or mediating mechanisms.