In this study, we observed an inverse association between adherence to the Healthy Eating pattern and overall, as well as high-grade PCa, with evidence of an exposure-response pattern. In contrast, adherence to the Western Sweet and Beverages pattern was associated with a higher PCa risk overall and for high-grade tumors. Increasing adherence to this pattern increased risk. No clear association was found in relation to the Western Salty and Alcohol pattern.
4.2. Methodological Considerations
Our study, including 1919 cases and 1991 controls, is the largest population-based case-control study to date to assess the role of dietary patterns among predominantly Caucasian men in PCa risk. It was conducted in Montreal, which harbors a dietary culture with a strong French influence. The dietary patterns identified here, as those in all other studies, reflect local cultural heritages, so perfect alignment of patterns across various study populations is not expected.
In studies of diet and chronic diseases, prospective designs are usually preferred, provided that that follow-up is long enough to allow for a sufficient number of cases to accrue. This may pose a challenge to the study of PCa, which generally develops at an advanced age, requiring prolonged follow-up. Unlike for case-control studies, which are subject to reporting bias, the prospective nature of previous cohort studies conducted on this issue alleviates concerns for recall bias based on disease status. However, in cohort studies investigating dietary patterns, assessments were typically conducted at study baseline among men of various ages, sometimes ranging as far apart as from ages in the 20s to the 80s, without accounting for changes in food consumption over time. The timing of the assessment in cohort studies may thus have reflected remote or recent intakes, depending on the study, whereas in all case-control studies, including ours, reports focused on recent diet, which may be more relevant to cancer progression. We were also able to capture changes in dietary intake of several key food groups within the 20 years preceding the index date. In analyses restricting subjects to those reporting no major changes in intakes over this period, results were not altered (data not shown).
There necessarily was some degree of misclassification of dietary exposures in our study. As in previous investigations, we used a validated FFQ, recognized as a standard method of dietary assessment for determining usual food intake [38
]. The questionnaire focused on the two years before diagnosis/interview to reduce the likelihood that cancer patients would report diets reflecting dietary changes post-diagnosis.
The majority of previous studies [12
], like ours, used PCA to derive dietary patterns. Others used exploratory factor analysis [23
] or pre-defined patterns such as eating indices [13
]. Our sample size was sufficiently large to use PCA with 72 dietary items which presented several advantages. PCA alleviates the problems of collinearity between the foods consumed concurrently [39
]. Moreover, it enables the use of continuous variables, which are estimated with FFQs, without having to create categorical variables, thereby reducing exposure misclassification.
In our study, the pattern explaining the highest percentage of variance of the food intake was Healthy Eating (8.6%), followed by Western Salty and Alcohol (7.2%) and Western Sweet and Beverages (3.7%), resulting in 19.5% of the total variance explained. This is similar to several other investigations where patterns explained around 11–38% of the total variance [12
], albeit less than in the study on nutrient-based patterns, explaining 78% of variance in nutrient intake [21
]. Of interest, the food items constituting the Health Eating pattern are aligned with the recommendations from Canada’s Food Guide [40
Participation rates were imperfect, albeit relatively good as compared to several previous investigations [41
]. Our comparison of participants and non-participants, separately for cases and controls, to census-derived variables indicated minimal differences between groups, consistent with the absence of a major selection bias. Selection bias based on food consumption is implausible as there was no mention to potential participants that the study included a dietary component.
Information on several covariates was available in the study, enabling their consideration when building our analytical models. Nevertheless, the potential for residual confounding or confounding by an unmeasured factor remains possible. For instance, physical activity may influence PCa risk indirectly through BMI or directly via various changes in other metabolic patterns not considered here [42
]. However, in a recent review of the association between physical activity and metabolic syndrome, physical activity was more often associated with a reduction in obesity (waist circumference) than with a resolution of other components of the syndrome [43
]. Furthermore, the American Institute of Cancer Research in their most recent update of the evidence for physical activity and prostate cancer risk has classified the evidence in the category of “limited – no conclusion” [5
While several statistical tests were conducted as part of this study, we did not implement corrections for multiple testing. The dietary patterns identified were derived using PCA and were not pre-defined based on a priori information, resulting in novel patterns specific to this study population and for which associations with PCa had never been evaluated. Therefore, given the hypothesis-generating nature of the study, it was judged preferable to accept the possibility that some of the associations observed might have occurred by chance, based on the premise that findings should be confirmed in future studies [44
Two large cohort studies suggest a protective effect of a “healthy or prudent” diet in PCa risk [13
]. This is the first case-control study to report a similar finding, in further support of the WHO’s recommendations of eating fruits and vegetables to reduce the risk of disease [46
]. Since they consider that foods are eaten together, and not alone, food patterns may be more easily amenable to health promotion interventions and chronic disease prevention.
Our findings document stronger associations, either protective (Healthy Eating) or harmful (Western Sweet and Beverages) with high-grade cancers, as compared to low-grade ones. Other studies also suggest associations that are specific to aggressive or advanced cancers, such as a lower risk with a Mediterranean pattern [15
] and a higher risk with a Western pattern [16
]. Ambrosini et al. found a higher risk among men with a Western pattern; results were attenuated for non-aggressive cancers [12
]. By contrast, Jackson et al. observed a higher risk with a Refined Carbohydrate pattern and results were more pronounced for low-grade PCa [17
]. There is evidence that less aggressive and more aggressive cancers may have different sets of risk factors and etiology [5
]. Indeed, low-grade and high-grade cancer foci progress largely in parallel, diverging early from a common progenitor [48
At the time our study was conducted, there was a very high uptake of PCa screening in the study base, despite the absence of a screening program. Screening was often integrated in routine yearly exams in this population with a free, universal access to healthcare. This distinguishes this study from many others, as detection issues can bias associations between exposures, including diet, and PCa [6
]. For instance, health-conscious individuals may tend to have both a healthier diet and undergo more closely medical follow-ups, including disease screening. Our high screening rate thus translates into a lower likelihood of bias by PCa detection than many other studies. Moreover, we had the ability to conduct a sensitivity analysis excluding controls who were not recently screened to reduce the potential of latent cases in our controls series. This had a minimal impact on our findings given the high proportions of screened controls. In a cohort study conducted by Shin et al., associations with dietary patterns differed when screening or subjective symptoms were considered [23