In this large prospective cohort of United States men and women aged 50 to 71, substantial inverse education gradients persist for incident cancer. In fully adjusted models, we found higher risks among the least, compared to the most, educated individuals, especially for combined smoking–related cancers (comprising those of the head and neck, esophagus, lung, bladder, and pancreas). In addition, we found inverse education gradients for cancers of the stomach and rectum (men only) and colon (women only). Some direct associations with education and cancer risk also emerged, notably those for melanoma of the skin (both men and women), localized prostate cancer (men), and invasive breast and endometrial cancer (women).
The NIH-AARP cohort is a large prospective cohort with detailed information on a variety of covariates which allowed us to control for multiple risk factors at the individual level in an analysis of first primary rare and common malignancies in both men and women. Other prospective studies conducted in Europe have reported similar results, although these studies did not control for risk factors
, analyzed only common cancers
, or presented data only for women.
Other studies in the U.S. have reported on the relation of education to cancer mortality, with results broadly similar to ours.
The availability of registry–based incidence data in our cohort focused the analysis on potential cancer causation, largely circumventing the complicating influence of treatment factors on cancer mortality outcomes.
The smoking–adjusted analyses are revealing in two ways. First, for some sites, particularly lung and smoking–related cancers combined, adjustment for smoking leads to substantial attenuation of the inverse education–cancer association in men and women. Given that smoking is clearly related to education () and smoking is an established cause of these cancers, this relative risk attenuation suggests strongly that smoking is a key intermediate factor on the education–cancer pathway. Second, although the education–cancer relative risks are attenuated by smoking adjustment, they do not revert to the null. Even after adjustment for smoking, the lung, esophageal and overall smoking–related cancer risks for the least, compared to the most, educated men remain approximately doubled. This may reflect residual confounding by smoking or the presence of causal factors other than smoking (be they biological or psycho–social) on the education–cancer pathway. That education, even after taking smoking and other factors into account, should consistently predict, for example, the development of esophageal cancer in men remains both tantalizing and a target for etiologic research.
After adjustment for age and smoking, the inclusion of other covariates in the regression models resulted in little additional attenuation of the education–cancer associations. Although residual confounding for such imperfectly measured variables as total energy intake, alcohol consumption, and physical activity cannot be ruled out, these additional factors explain relatively little of the education–cancer connection.
We did not have information on H. pylori infection status to incorporate in the multivariate analyses of gastric cancer. However, when investigated by Nagel et al in a large nested case control study in Europe, the inverse association of gastric cancer remained, albeit non-significant, even after controlling for H. pylori
The data reveal a strong inverse gradient for pleural cancer in men. This finding from a prospective cohort study, possible only because of the study's large size, appears unexplained by smoking and may reflect occupational or environmental exposure to asbestos.
It is noteworthy that asbestos was used widely in the United States until the implementation of the Occupational Safety and Health Administration (OSHA) regulations in 1971, when the study participants were approximately aged 26–47 and thus of sufficient age to have accrued occupational or environmental exposure.
Education level was weakly but significantly positively associated with invasive breast cancers in women, which is consistent with findings from other studies.
Age at first birth, parity, and use of MHT are all related both to education and breast cancer, which likely accounts for the modest attenuation of the positive education–breast cancer relation in the multivariate analyses. In contrast to some other studies, endometrial cancer was directly related to educational attainment and this association was not attenuated after adjustment for BMI and MHT in the multivariate analyses.
The modest overall positive association between all cancer incidence and educational attainment appears to be largely driven by the positive associations for breast and endometrial malignancies.
Studies of educational attainment and prostate cancer have yielded inconsistent results. In our cohort, the education–prostate cancer association was weakly positive, statistically significantly so only for localized disease. The point estimates were similar for localized and advanced prostate cancer, however; the power to detect the positive association with advanced disease was limited. The weak positive association for prostate cancer was largely unaffected by multivariate analysis, which is not surprising given the paucity of strong risk factors for this malignancy.
The direct association of education level with melanoma of the skin in our cohort is in line with previous findings. 
In general, higher SES individuals are more likely to participate in outdoor leisure activities and vacation in places with high sun exposure
, and for this reason may have increased melanoma risk.
It is important to note that the AARP membership tends to be more educated, on the average, than the U.S. population as a whole. Nevertheless, the cohort has a wide range of educational attainment, including over 30,000 people, or 6.6% of the study population, with less than a high school education. This wide range of educational attainment allows us to make informative comparisons of cancer incidence across education categories.
Education captures many aspects of the constructs ‘social class’ and ‘socioeconomic status’ and is widely used as an indicator of social ‘difference’ in epidemiologic studies. A particular advantage of investigating education is avoiding reverse causation bias: incident cancer may lead to downward occupational mobility and reduced income but generally will not affect educational status achieved by early adulthood.
In summary, the data from the NIH–AARP cohort show that substantial education gradients in incident cancer risk persist in the United States. A few malignancies are positively associated with educational attainment; these positive associations are primarily of etiologic interest, given that lowering educational attainment is hardly an appropriate strategy for preventing melanoma of the skin or cancers of the breast, prostate, and endometrium. The majority of the observed education associations, however, are inverse, and these are evident especially for smoking– related malignancies. Smoking likely accounts for some—although not all––of the increased cancer risk among lower educated men and women. To the extent that smoking is the mediating causal factor, reducing the differential in smoking rates is a reasonable strategy for addressing SES–cancer inequalities. To the extent that smoking does not account for the inverse associations, further research to identify the causal factors underlying the education–cancer gradients is clearly warranted.
The persistent education-cancer differences in the United States (and many other countries) remain a cause for concern. They also, however, present an opportunity to understand more deeply the etiology of cancer and ultimately reduce its incidence.