Associations between an obesity-related dietary pattern and incidence of overall and site-specific cancers: a prospective cohort study

Cancer produces poor health outcomes and imposes a heavy burden on society. Globally, cancer was responsible for 9.6 million deaths and 18.1 million cases in 2018 [1]. Therefore, there is an urgent need to reduce cancer incidence by altering modifiable risk factors, such as diet, one of the established determinants for several site-specific cancer [2]. Specific unhealthy dietary factors were identified as risk factors for corresponding cancer sites by the World Cancer Research Fund (WCRF) Third Expert Report [2]. Therefore, the relationship between diet and multiple site-specific cancer has been a persistent concern by the general population and researchers. Although numerous studies have investigated isolated nutrients and food, the health effect of individual dietary factors on cancer is thought to be limited and equivocal. By contrast, the exploration of dietary patterns (DPs) may yield stronger cancer-related health effects and more reliable estimations, so the results can be more readily applied on clinical practice and dietary guidelines [3, 4]. DPs can be developed based on investigator-defined criteria or data-driven analyses. For instance, we can classify food items into several subsets and assign them scores based on prior knowledge, or we can extract the DP by a data-driven technique such as reduced rank regression (RRR), which derives DPs with the help of the given response variables. Compared to methods such as principal component analysis and index-based methods, RRR combines prior knowledge with a posteriori data-driven technique. RRR uses disease-related information (prior knowledge) to select response variables, which are pre-hypothetical intermediates between DPs and diseases, and then empirically derives DPs (posterior data-driven technique) to explain the maximum variations in response variables. Therefore, it is more likely to develop DPs related to the given health outcomes. Moreover, RRR can test a putative hypothesis of disease pathophysiology [5], which is helpful for mechanistic evidence. In sum, we need to select a set of response variables on the pathway between diets and cancer to derive the DP.

Indeed, diets play various roles in different pathways such as weight gain [6], apoptosis [7], oxidative DNA damage [8], and interplay with gut microbiota [9]. Among these pathways, obesity, a pandemic metabolic disorder disease, is a vital and modifiable mediator between diets and cancer that rose from nearly 20% to 40% among adults from 1975 to 2016 [10]. The associations between obesity and overall and site-specific cancers have been validated by a mendelian randomization study for the UK Biobank [11] and other large cohort studies [12, 13]. Moreover, obesity can be evaluated in a non-invasive and low-cost way compared to other metabolic syndrome (MetS) components (except for blood pressure), which are assessed by blood samples. Hence, proxy indicators of obesity were the ideal response variables to derive the obesity-related DP for this study. Therefore, in the present study, we first developed an obesity-related DP employing RRR and then extensively examined the prospective associations of the obesity-related DP with overall cancer and 19 site-specific cancers in the general UK population.

The first DP explained the largest amount (11.2%) of variation in the response variables and showed positively strong correlations with WHR (r = 0.80) and BMI (r = 0.60) (Additional file 1: Table S4). As shown in Fig. 1, the obesity-related DP is characterized by a higher intake of beer and cider, processed meat, high sugar beverages, red meat, and artificial sweetener, and a lower intake of fresh vegetables, olive oil, tea, and high fiber breakfast cereals. The obesity-related DP Z-scores range from -3.97 to 7.27.

We included 114,289 participants, among whom 10,145 (8.9%) had incident cancers. The median follow-up period was 9.4 years for overall cancer. Baseline characteristics across quartiles of obesity-related DP Z-scores are displayed in Table 1. Individuals in the highest quartile tend to be male, less educated, current smokers, physical inactive, and higher in energy intake. Additionally, they tend to have worse health conditions, such as hypertension, diabetes, high cholesterol, and CVD at baseline (all P-values < 0.001).

In this prospective study, we developed an obesity-related DP, which was characterized by a higher intake of beer and cider, processed meat, high sugar beverages, red meat, and artificial sweetener, and a lower intake of fresh vegetables, olive oil, tea, and high fiber breakfast cereals. General individuals in the highest versus the lowest obesity-related DP quartile were at elevated risk of overall cancer and oral, colorectal, liver, lung, endometrial, ovarian, kidney, and thyroid cancers. Moreover, the obesity-related DP was associated with different cancer sites among four subpopulations grouped by a combination of age groups and sexes. Additionally, we validated that the association between obesity-related DP and overall cancer was mainly mediated by the BMI and WHR (two response variables) and along with C-reactive protein and HDL and triglycerides. These findings have meaningful clinical and public health implications in the prevention of cancers through dietary approaches among the UK population.

From the multi-pathways between DPs and cancer, we chose one of the most effective pathways by setting the proxy indicators of obesity as the inter-mediator to bridge the indirect relationship between the obesity-related DP and cancer because obesity, an important established risk factor of several site-specific cancers [12], is the second biggest risk factor of cancers in the UK [25], following smoking. Although the variations of nutrients as more proximal response variables versus biomarkers can be substantially explained, it may violate the independence assumption for response variables when using predictors (food groups) and response variables from dietary assessment tools simultaneously [26]. Compared to 11.2% variation in response variables of our study, we noticed that previous studies also displayed low variations ranging from 1.7% to 4.2% [27‐29] using biomarkers as response variables for RRR. Previous studies showed consistent results on certain cancer sites. For example, a meta-analysis of 93 studies suggested that prudent or healthy DPs are associated with a decreased risk of breast, colorectal, and lung cancer [30]. Additionally, the Western diet has been associated with a higher risk of colorectal [31, 32], breast [33, 34], esophageal [35], pancreatic [36], ovarian [37], and prostate cancers [38]. Likewise, we found that obesity-related DP was associated with overall and site-specific cancers in the digestive system. Moreover, we detected that certain non-digestive cancer sites, which have been less reported on, such as lung, kidney, bladder, endometrium, ovary, malignant melanoma, and multiple myeloma, were linked to the obesity-related DP. In total, we provided clues about potential relationships between the obesity-related DP and various cancer sites based on the given pathways.

Our study also systematically assessed the role of the obesity-related DP in the progress of a wide range of cancer sites. In the present study, beer and cider were in the unhealthy food group with the highest factor loading. A similar analysis between drink types and diseases also showed that beer and cider consumption was associated with a 14% higher risk of cancer and increased risks of CVD and all-cause mortality in the general UK population after excluding non-drinkers [39]. Red and processed meats, important components of the obesity-related DP and the Western diet, also play nonnegligible roles in carcinogenesis. An umbrella review summarized 72 meta-analyses and established that red and processed meat were separately associated with nine and 10 types of cancer, respectively, which was the same as our findings at specific cancer sites (colorectal, endometrium, esophagus, and lung) [40]. With urbanization and economic growth, sugar-sweetened and artificially sweetened beverages, major sources of added sugars in the diet, have become more popular in many populations, but they cause weight gain and an increased risk of cancer and other chronic diseases. Moreover, the adverse effects of sugar-based beverages on health outcomes, which have been shown by numerous cohort studies and clinical trials, were also demonstrated by Malik et al. [41]. Moreover, a meta-analysis selected 27 of 64 studies and found that sugar-sweetened beverages were positively associated with breast and prostate cancer risk and tended to be associated with colorectal and pancreatic cancer risk [42], which is partly in line with the results of our study. As to health factors in the obesity-related DP in this study, fresh vegetables and olive oil were major contributors to higher negative factor loadings, which were related to a lower extent of obesity. The Mediterranean diet also recommends higher intakes of fresh vegetables and olive oil, which similar studies have established to be related to a lower risk of colorectal cancer (RR _{observational}: 0.82, 95% CI: 0.75 to 0.88; n = 11 studies) [43]. Further, a higher intake of vegetables has been found to decrease the risk of lung cancer by 8% (n = 25 studies) [44] and overall cancer by 10% (n = 15 studies) [45]. From a biochemical perspective, the chemopreventive activity of phenolic compounds in olive oil demonstrated an ability to inhibit oxidative DNA damage in several human and animal models [46], and other evidence also suggests the favorable effects of phenolic compounds on free radicals, inflammation, and carcinogenesis [47, 48]. Moreover, the beneficial effects of fiber and tea, both high in negative factor loading on cancers, have also been observed. Dietary fiber improves glucose and lipid parameters via the short-chain fatty acids produced by microbial fermentation [49] and inhibits the carcinogenic effects of N-nitroso compounds by acting as a nitrite scavenger [50]. Tea is one of the most popular beverages worldwide, and the protective effect of tea on six cancer sites has also been observed [51]. Tea is rich in various polyphenols, which may inhibit tumor formation and growth through its antioxidative and potentially antiproliferative roles [52]. Moreover, the consumption of alcohol in beer and cider generates the ethanol's metabolite acetaldehyde in blood circulation causing DNA damage and block DNA synthesis and repair [53]. Meat is also rich in harmful components such as heterocyclic amines, polycyclic aromatic hydrocarbons, and nitrate, which can impact the development of cancer [40]. Taken together, the compositions and contributions of the obesity-related DP have demonstrated a reasonable impact on the onset of cancers, but previous studies have shown borderline associations. In contrast, we developed an obesity-related DP, which was more strongly and widely associated with overall cancer and multiple site-specific cancers.

We observed different nonlinear association patterns for site-specific cancers; the inflection points were consistently at around 0 or 2 in terms of obesity-related DP Z-scores. The “platform-to-increase” risk for esophageal cancer might be explained by the previous evidence that suggests the opposite effects of BMI on the esophagus cancer type (adenocarcinoma versus squamous cell carcinoma) [54], whose constituent ratio might result in the J-shaped association. The “fast-to-low decrease” relationship for malignant melanoma cancer might reflect a real nonlinear biological association or partly be attributed to the disproportional contributions of the obesity-related DP to body fat-free mass and fat mass, whose opposite effects (i.e., adverse vs. protective) on malignant melanoma cancer were suggested by a study in UKB [55]. The inverse U-shaped relationship for prostate cancer might be caused by delayed or missed diagnoses in people with a high BMI, which is supported by the finding of the opposite effects of BMI on localized and advanced prostate cancer [56]. Moreover, the slight protective effect of the obesity-related DP on bladder cancer at the lower scale of the obesity-related DP Z-score could be attributed to heterogeneity across smoking status because the opposite effects were found among nonsmokers (protective) and current/previous (adverse) smokers (Additional file 1: Table S8, P for interaction < 0.001). Similarly, heterogeneity across sex for multiple myeloma cancer in the opposite effects of DP Z-score might explain the inverse U-shaped association. These heterogeneities indicate that there are different mechanisms or combinations of mechanisms associated with different cancer sites among different subgroups. Moreover, another observational study in UKB suggested different patterns of nonlinear associations between C-reactive protein (potential mediator) and cancer sites such as the kidney [23]. Taken together, the pooled effect of mediators might partly explain these complex nonlinear relationships, but the underlying mechanisms must be explored further.

Obesity is not merely an excess of adipose tissue; it also accompanies metabolic dysfunction (hyperglycemia, hypertension, and dyslipidemia) and local adipose tissue inflammation [57]. Indeed, the unfavorable effects of obesity on cancer development and progression are attributable to disruptions in adipokines, sex hormones, inflammation, and insulin metabolism [58]. Likewise, we found that the obesity-related DP could play a role in related inflammation pathways because C-reactive protein, one of the proxy indicators of inflammation, was the potential mediator in this study. This indicates that the obesity-related DP induces a local or systematic inflammation environment in the body, along with excess and abnormal fat accumulations. Moreover, HDL, as another mediator, partly bridged the connection between the obesity-related DP and overall cancer. As we know, HDL plays a key role in regulating cholesterol as well as anti-inflammation, antioxidation, and immune modulation, which is important in cancer incidence. Additionally, there is an assumption that HDL may facilitate the exchange of cholesterol between healthy and cancer cells [59].

In the present study, we used two or more dietary assessments to reflect standard intakes. Then, we constructed all of the models weighted by the number of dietary assessments to make our results reliable. Moreover, we set the last dietary assessment date as baseline time rather than the date of recruitment or first dietary assessment to avoid reverse causality. However, the insufficient number of cases in some cancer sites (oral, cervix, and thyroid) may have led to statistical inefficiency owing to the strict inclusion criteria in this study. Another limitation of our study was that dietary intakes were measured by multiple 24-h online dietary assessments, which were dependent on self-reporting, which may have caused misreporting and recall bias. To reflect the standard intake and capture reliable estimates, we used dietary data from a minimum of two 24-h online dietary assessments to derive the obesity-related DP, and then we compared it with the DP derived from distinct times dietary questionnaires to validate the stability of the obesity-related DP. Moreover, the response variables and dietary intakes might change over time. However, our study did not reflect long-term changes in dietary intake and body weight. However, we used response variables measured at an intermediate time point to validate the robustness of the results. Because this study was observational and confounded by potential unadjusted confounding factors, it is premature to infer causality. Finally, our posterior data-driven approach and regional dietary cultures make it difficult to generalize our results to other populations.

Overall, we found that the obesity-related DP was strongly associated with a higher risk of overall cancer and multiple site-specific cancers. In general, in UK adults, adherence to the obesity-related DP was linearly associated with an increased risk of overall cancer and three site-specific cancers (oral, colorectal, and liver) in the digestive system and three site-specific cancers outside of the digestive system, and it was nonlinearly associated with six site-specific cancer sites. In the future, further studies with longer follow-up times and larger sample sizes will be conducted to identify other similar DPs.