Educational level and risk of colorectal cancer in EPIC with specific reference to tumor location



Existing evidence is inconclusive on whether socioeconomic status (SES) and educational inequalities influence colorectal cancer (CRC) risk, and whether low or high SES/educational level is associated with developing CRC. The aim of our study was to investigate the relationship between educational level and CRC. We studied data from 400,510 participants in the EPIC (European Prospective Investigation into Cancer and Nutrition) study, of whom 2,447 developed CRC (colon: 1,551, rectum: 896, mean follow-up 8.3 years). Cox proportional hazard regression analysis stratified by age, gender and center, and adjusted for potential confounders were used to estimate hazard ratios (HR) and 95% confidence intervals (95%CI). Relative indices of inequality (RII) for education were estimated using Cox regression models. We conducted separate analyses for tumor location, gender and geographical region. Compared with participants with college/university education, participants with vocational secondary education or less had a nonsignificantly lower risk of developing CRC. When further stratified for tumor location, adjusted risk estimates for the proximal colon were statistically significant for primary education or less (HR 0.73, 95%CI 0.57–0.94) and for vocational secondary education (HR 0.76, 95%CI 0.58–0.98). The inverse association between low education and CRC risk was particularly found in women and Southern Europe. These associations were statistically significant for CRC, for colon cancer and for proximal colon cancer. In conclusion, CRC risk, especially in the proximal colon, is lower in subjects with a lower educational level compared to those with a higher educational level. This association is most pronounced in women and Southern Europe.

Colorectal cancer (CRC) accounts worldwide for about 1 million new cases per year and it is the fourth most common cancer in men and third most common in women.1 Worldwide, the incidence of CRC varies at least 25-fold, with the highest occurrence in North-America, Australia/New Zealand, Western Europe, and, in men especially, Japan.1 This variation in incidence throughout the world and the increased incidence of CRC over the last 40 years suggests an influence of lifestyle and environment over genetic factors.1–3

Several studies have been performed to examine the relationship between socio-economic status (SES) and CRC incidence. A British cohort study4 found no association between low SES, social deprivation, and risk of CRC. Another British study performed in 2003 also observed no significant impact of socio-economic deprivation on CRC incidence.5

A large Italian multicenter case-control study from 1999 including 3,533 cases observed however that a higher educational level and a higher social class increased the risk of cancer in the colon, but not the rectum.6 This was thought to be due to known risk factors for CRC, including lower occupational physical activity and dietary factors such as a higher intake of meat and eggs in the higher social class.6 Similar results were found in a Dutch cohort-nested case-control study performed in 1995,7 in which a positive association was found for cancer in the colon in the highest education levels. A large cohort study conducted in Finland reported higher standardized incidence ratios for cancer in both the colon and rectum in higher social classes.8

In contrast, according to a large cohort study9 including participants from the United States (US), low SES was associated with an increased risk of CRC, which was only partially caused by smoking. Another cohort study conducted in the US and Canada reported a higher CRC risk in participants with lower income.10 Similarly, a systematic review from 2005,11 which included a total of 46 studies with data from participants living in the US, reported an increased CRC risk in participants with a low SES. These results were confirmed in a recent systematic review,12 which reported an increased CRC incidence among participants with a low SES in the US. In addition, US data have reported a higher CRC incidence and presentation at a later disease stage in the black population.11 Higher mortality rates in this group are almost completely explained by lower SES in this population.13 Lower health care utilization seems to be an important factor in these observations, since studies conducted within the Veteran's administration, where access to care can be expected to be similar to all veterans, showed no racial differences in treatment and survival.11

Besides variations between geographical regions, other reasons for discrepancies between study results could be that not all authors adjusted their analyses for confounders or investigated risk factors that could explain the observed associations. Further insight in whether CRC risk is SES-specific could therefore be of interest; as a statistically significant result despite extensive correction could point to SES-specific variables causing a higher CRC risk.

In this study, we investigated the association between SES, as defined by the highest level of education, and risk of CRC in participants of the European Prospective Investigation into Cancer and Nutrition (EPIC), a large prospective European cohort study, in which an adjustment for a wide range of variables can be made.


95%CI: 95% confidence intervals; BMI: Body Mass Index; CRC: Colorectal cancer; EPIC: European Prospective Investigation into Cancer and Nutrition; HR: Hazard ratios; RII: Relative indices of inequality; SAS: Statistical analyses system; SES: Socio-economic status; US: United States; FFQs: food frequency questionnaires

Materials and Methods


EPIC is a multicenter prospective cohort study designed to investigate the relationship between diet, nutritional and metabolic characteristics, lifestyle factors and the risk of cancer. Enrolment of subjects took place between 1991 and 2000 in 23 collaborating centers in 10 European countries (Denmark, France, Germany, Greece, Italy, The Netherlands, Norway, Spain, Sweden and the United Kingdom). Participants were usually selected from the general population of a specific geographical area and the vast majority of participants were Caucasian. The selection of the study-population, collection of data and procedures for follow-up were previously reported in detail.14–16 In brief, standardized lifestyle questionnaires, food frequency questionnaires (FFQ) and country-specific dietary questionnaires were used, blood samples were collected and anthropometric measurements were performed.14, 15

In total 521,448 participants, ∼70% women, mostly aged 35-70 years, participated. Of these, 3,100 subjects were identified as first incident CRC cases. All participants signed an informed consent form before enrolment. Participants were excluded in case of any prevalent cancer at recruitment (n = 23,523) or when information on follow-up time (n = 3,558) or the occurrence of a first primary colorectal tumor (n = 14) was missing. We also excluded participants with carcinoma in situ or unknown morphology (n = 92), participants with lowest and highest 1% of the ratio of energy intake to estimated energy requirement (n = 9,670) or participants with missing information on nondietary data (n = 1,221), or missing or nonspecified information on education (n = 6,643 and n = 10,942, respectively). Finally, all participants from the French cohort were excluded since this population was demographically homogeneous, consisting of women mostly concentrated in the two highest educational levels (n = 65,275, including 337 first incident cancers of the colorectum).

The current analyses were conducted on the data of 400,510 participants, including 2,447 CRC-cases.

Outcome assessment

Participants in the EPIC cohort are followed for cancer incidence and total mortality. In most of the participating centers, the identification of cancer cases is based on population cancer registries (Denmark, Italy, The Netherlands, Norway, Spain, Sweden and the United Kingdom). In Germany and Greece a combination of methods including health insurance records, cancer and pathology registries and active follow-up through participants or relatives is used.15 Specific censoring dates were established depending on the dates in which the cancer registries were considered to be complete, i.e., between 2002 (Granada, Spain) and 2006 (Greece).

The 2nd edition of the International Classification of Diseases-Oncology (ICD-O) was used to code the cancer incidence data. Cases included all first primary incident CRCs. Tumors were subdivided for location, i.e., rectum and colon, and then further subdivided. Proximal colon tumors included the cecum, appendix, ascending colon, hepatic flexure, transverse colon, and splenic flexure (C18.0–18.5). Distal colon tumors included the descending (C18.6) and sigmoid colon (C18.7). Overlapping (C18.8) and unspecified lesions (C18.9) of the colon were grouped among colon cancer only (C18.0–C18.9). Cancer confined to the rectum included tumors occurring at the rectosigmoid junction (C19) and rectum (C20).

Statistical methods

Highest educational level attained was used as a proxy for SES. This information was collected using a questionnaire specific for each country and the replies were then categorized into a common classification using four categories (primary education or less, vocational secondary education, other secondary education, college or university).

Baseline characteristics were presented according to educational level using means with standard deviations, medians with interquartile ranges or frequencies, whenever appropriate.

Analyses were conducted using Cox regression models. We computed hazard ratios (HR) by including participants with college or university as reference category. Age was used as the primary time metric. Time at entry was age at recruitment and exit time was the age when participants were diagnosed with cancer, died, emigrated, lost to follow-up or censored at the end of the follow-up period, whichever came first. To control for differences in questionnaires, follow-up procedures, and other center effects, all analyses were stratified by center. The results were then further stratified by age at recruitment in 1-year categories and by gender.

We also computed relative indices of inequality (RII). The calculation of the RII is based on a relative measure of education.17 Each individual is assigned a fractional rank (from 0 to 1) corresponding to the mean proportion of the population with a lower level of education, using the mean rank for ties. The RII is then computed with a Cox regression model using this ranked variable as independent variable. Thus, the RII expresses inequality within the whole socioeconomic continuum. It can be interpreted as the ratio of the expected CRC risk between the highest and the lowest (0th–100th percentile rank) educated participants in the population. The RII was used to quantify the differences in incidence rates that are systematically related to a higher or lower educational level. Even if parts of the relationship are nonlinear, we were primarily interested to know whether there is a relationship to educational level. For such a purpose, the RII is a very adequate measure because it separates the linear from the nonlinear gradients. As the RII considers size and relative position of each educational group, it minimizes problems due to differences between countries in the educational classification and distribution. This ranked variable was computed by age category (age at recruitment; <30, 30–40, 40–50, 50–60, ≥60) and center.

Univariate and multivariate analyses were performed. The variables that were considered were Body Mass Index (BMI) (kg/m2), Cambridge physical activity index (inactive, moderately inactive, moderately active, active),18 alcohol consumption (g/day, estimated average alcohol intake at recruitment), fruit and vegetable consumption, fiber intake, energy intake from fat and nonfat, consumption of red meat, processed meat and fish, all the former in g/day, smoking status (never, former, current) and smoking duration (years). BMI was chosen over height and weight because of a greater effect on risk estimates. Information on lifetime number of cigarettes smoked and waist circumference was not available for a large number of participants. Furthermore, in sub-analyses these variables did not markedly change the risk estimates and were therefore not added to the fully adjusted model. We conducted analyses among all participants and performed stratified analyses by gender, geographical regions and age groups.

Presence of interaction on a multiplicative scale by gender, age group and geographical region was tested using the interaction term of educational level with gender (male; female), age group (≤60 years, >60 years) or geographical region (Northern Europe: Norway, Sweden and Denmark. Middle Europe: United Kingdom, The Netherlands and Germany. Southern Europe: Spain, Italy, Greece).

All statistical analyses were performed using the statistical analyses system (SAS) software package, version 9.2 (SAS institute, Cary). Two-sided p-values < 0.05 were considered statistically significant.


Our analyses were based on 2,447 CRC cases (48.3% men) and 398,063 noncases (35.0% men). Of all cases, 896 were rectum cancers (54.5% men) and 1,551 colon cancers (44.7% men), of which 687 were in the proximal colon (42.1% men), 703 in the distal colon (46.5% men) and 161 cancers were not further specified for location (47.8% men). The mean follow-up time was 8.3 years. Table 1 shows the CRC cases and total cohort stratified by country. All levels of education were well represented in the countries. The highest percentage of participants with the lowest educational level was reported in Spain, whereas the highest percentage of participants with the highest education level was reported in the United Kingdom.

Table 1. Description of colorectal cancer cases and cohort by country
inline image

Baseline characteristics stratified by educational level are shown in Table 2. Compared with participants from lower education categories, participants from higher education categories were taller, had a slightly lower BMI, were less likely to be current smoker, had a shorter duration of smoking, a higher alcohol and fiber intake, a lower consumption of red meat and fish, and were less inactive, but were more likely to be moderately inactive or moderately active.

Table 2. Baseline characteristics according to Socio-economic status1
inline image

Table 3 shows the number of cases and unadjusted and fully adjusted HRs. Full correction strengthened the inverse risk estimates. In the colorectum overall, and in the colon and rectum separately, nonsignificantly decreased adjusted risks for CRC were seen in participants with vocational secondary education or less, compared to participants with a college or university education. The adjusted risk for cancer in the proximal colon was statistically significantly decreased for participants with a vocational secondary education (HR 0.76, 95% CI 0.58–0.98) and a primary education or less (HR 0.73, 95% CI 0.57–0.94). In contrast, for the distal colon nonsignificantly increased adjusted risks were seen for all educational levels compared to participants with a college or university education.

Table 3. SES and risk of colorectal cancer by site, unadjusted and fully adjusted hazard ratios1
inline image

Table 4 shows the effect of known risk factors on RIIs by adjusting on a one-by-one basis. None of the variables alone fully explained the decreased risk of proximal colon cancer in participants with a lower educational level. Full correction further strengthened all inverse risk estimates with the one for CRC becoming borderline statistically significant (RII 0.83, 95%CI 0.70–1.00) and the one for proximal colon cancer only becoming statistically significant (RII 0.65, 95%CI 0.47–0.91).

Table 4. SES and risk of colorectal cancer by site, relative indices of inequality1,2
inline image

For CRC, interaction was found with gender (p = 0.01), age group (p < 0.01) and geographical region (p < 0.001). In subgroups of gender, most RIIs showed a decreased cancer risk for participants with a lower education compared with participants with a higher education, in both men and women (Table 5). In both genders, fully adjusted risk estimates were lowest for proximal colon cancer. Only in women these decreased adjusted risks were statistically significant for CRC overall (RII 0.75, 95%CI 0.59–0.96), for CRC confined to the colon (RII 0.69, 95%CI 0.51–0.93) and for proximal colon cancer (RII 0.60, 95%CI 0.39–0.93).

Table 5. SES and risk of colorectal cancer by site, unadjusted and fully adjusted relative indices of inequality, stratified in subgroups of gender, age and region1
inline image

Except for CRC confined to the rectum, RIIs stratified for geographical region showed a north-south gradient with lowest risks for Southern Europe. Only for Southern Europe adjusted risk estimates were statistically significantly decreased (RII 0.60, 95% CI 0.42–0.86 for CRC, RII 0.50, 95% CI 0.33–0.77 for CRC confined to the colon and RII 0.26, 95% CI 0.13–0.52 for proximal colon cancer), indicating a decreased cancer risk for participants with a lower education in Southern Europe.


After extensive adjustment for risk factors, a low educational level compared to a higher educational level was found to be associated with a decreased CRC risk, especially in the proximal colon. This inverse association was most pronounced in women and in Southern Europe. Our results are consistent with a recent systematic review,12 which reported an association between an increased CRC incidence rate and higher SES among participants in Europe.

People with lower SES have been shown to be more likely to present with a more advanced CRC stage compared to people with a higher SES.19, 20 This is most likely due to a delayed diagnosis; however, a more aggressive growth pattern of CRC in people with a lower SES cannot be excluded. Some CRC cases in the lower educational categories might therefore not have been captured during follow-up, and consequently the decreased risk among lower educated participants might be (at least partially) due to lead time bias. An analysis on mortality, which may be less affected by lead time bias, might shed further light on this issue. Whether lead time bias is an issue in our study could also be assessed by looking at tumor stage at diagnosis, however unfortunately we do not have any reliable information on tumor stage in our database.

Our finding of an effect mainly in the proximal colon, and not in the distal colon and rectum, is another indication that right- and left-sided CRC cancer are separate entities that are affected by different risk factors.21, 22 A previous prospective cohort study23 concluded that the risk of CRC confined to the colon was more influenced by known CRC risk factors than that of CRC confined to the rectum. Furthermore, in line with these findings are the results of a large multicenter case-control study in Italy,6 which reported an increased cancer risk in the colon for participants with more educational years, whereas such association was not found in the rectum.

To our knowledge, an inverse association between the occurrence of CRC and a lower educational level among women has not been reported previously. We have no clear explanation for this finding since full correction for known risk factors even further reduces the risk estimate, which could also be due to chance.

We did not observe an association between educational level and CRC risk in Northern and Middle Europe. For Northern Europe, this is in line with a cohort study that was previously conducted in Norway, which also showed no association between SES and CRC.24 The association between CRC risk and lower education in Southern Europe may possibly be linked to an increased prevalence of lower educated participants living in rural areas adhering to the Mediterranean diet.25 They might for example have a higher consumption of home grown grains, fruits and vegetables that was not sufficiently captured by the dietary factors we entered into our models. However our findings might also be due to chance.

Strengths of our study are its prospective design with the advantage that all information on education and potential confounders was collected before the occurrence of CRC. The large number of CRC cases with a considerable period of follow-up enabled us to perform sub-analyses for tumor localization, age, gender and geographical region.

Educational level was used as a measure for SES, which is a frequently used indicator for SES.26 It allows for classification of all individuals, including those not working or being retired. We cannot rule out possible inconsistencies between centers, especially since educational systems vary across European countries. However, a common classification of education level was used in all centers. The computation of the RII assumes a hierarchical order between all educational categories, but the ranking between the categories “vocational secondary education” and “other secondary education” is not always straightforward.27 However, we think that these limitations most likely have not materially affected the risk patterns in our population.

Second, our analyses were controlled for well known risk factors of CRC, such as BMI, physical activity, smoking, alcohol, dietary fiber and red meat intake. As none of the variables alone fully explained the relationship between educational level and CRC, we hypothesize that there is not just a single risk factor, but rather a combination of several risk factors that plays a role in educational differences in CRC risk. Nevertheless, since most of these factors are measured with a certain degree of error, residual confounding could still be a possible explanation for our findings.

A limitation of our study is that we assessed only one socioeconomic indicator, i.e., educational level, since information on occupational status at recruitment of the subjects and their spouses was not available for a sufficient number of participants and data on income was not collected. Multiple SES-related variables may be necessary to fully investigate the association between SES and CRC.28

Additionally, information on diet measurements was collected with food frequency questionnaires (FFQs). FFQs are indeed prone to measurement error,29 but this methodology is the usual approach in large population studies.

Participation in CRC screening programs is related to educational level, as patients with lower educational level are less likely to have participated in screening.30 This may have influenced our results. Additionally, although screening activities are distributed throughout the whole of Europe (national screening programs: Germany and the UK, regional screening programs: Denmark, Finland, Italy, Spain and Sweden31, 32), we cannot exclude the possibility that differential screening practices may explain our findings on geographical regions. Unfortunately, we do not have any specific information on screening participation in our cohort.

In conclusion, after extensive adjustment for known risk factors, CRC risk, especially in the proximal colon, remains inversely associated with a lower educational level. In our database, this association is most pronounced in women and in participants from Southern Europe. Although we cannot rule out that our findings are caused by chance, possible explanations could be lead time bias among participants with a lower educational level and uncontrolled effects of dietary and other lifestyle factors.