This systematic review and meta-analysis of cohort studies found a reduced risk of total cancer (13%), stomach (45%), colorectal (14%) and subsites (colon; 21%, and proximal colon; 45%), pancreatic (23%), breast (8%) and postmenopausal breast (19%), and bladder (21%) cancers, melanoma (21%), and non-Hodgkin’s lymphoma (24%) among vegetarians when compared to non-vegetarians. Suggestive non-statistically significant inverse associations were observed for several other cancer sites (lung, ovarian and prostate cancers), but these associations need further investigation in future studies. In addition, vegan diets were associated with a reduced risk of total cancer (23%) and breast cancer (20%), but no association was observed for colorectal and prostate cancer. When using alternative estimates from the Adventist Health Study II at age 65, the reduction in breast cancer risk was slightly strengthened (24%) and a reduced risk was also observed for prostate cancer (42%) among vegans.
The current findings are consistent with a few previous meta-analyses that reported reduced total cancer incidence in vegetarians and vegans vs. non-vegetarians [39,40,41], but is to our knowledge the first to analyse all specific cancer sites with available data and to report a reduction in risk of seven specific cancer types in addition to total cancer among vegetarians vs. non-vegetarians.
Several potential mechanisms could explain the observed associations between vegetarian and vegan diets and reduced cancer incidence. Differences in body weight could contribute to the observed associations as adiposity is an important risk factor for at least 12 different cancer sites [5], and possibly several additional cancers [42]. Vegetarian and vegan diets have been shown in randomized controlled trials to lead to weight loss when compared to non-vegetarian diets, with a mean difference of around 3.4 kg reported in one meta-analysis [31], and long-term cohort studies have reported increased weight gain over time with higher intake of red and processed meat and poultry and lower intakes of fruits, vegetables, whole grains and nuts [43,44,45]. Some of the observed associations between vegetarian and vegan diets and cancer risk were slightly stronger when not adjusted for BMI compared to when adjusted for BMI, and suggested differences in baseline adiposity explains 22% and 42% of the observed reduction in overall cancer risk among vegetarians and vegans, respectively, however, the proportion explained by BMI differed across cancer sites.
There is evidence of reduced risk of type 2 diabetes in vegetarians [46], and type 2 diabetes is an established risk factor for at least 6 cancer types [47]. There is also evidence that inflammation [48] and hormonal factors [49] plays an important role in cancer development. Several studies and reviews found vegetarian diets were associated with reduced insulin resistance and fasting insulin [50], lower C-reactive protein, fibrinogen and leukocytes [51, 52], and lower blood concentrations or urinary levels of estradiol and higher levels of sex-hormone binding globulin (SHBG) [53,54,55,56,57]. An intervention study using a mostly plant-based diet reported significantly increased SHBG, a borderline significant reduction in free estradiol, and reduced insulin, fasting glucose, C-peptide, growth-hormone binding protein, IGFBP-1, and IGFBP-2 [58]. Differences have also been reported in the microbiota of vegetarians compared to non-vegetarians [59,60,61,62] that may be beneficial in relation to cancer risk [63, 64].
Dietary differences between vegetarians and non-vegetarians are likely to contribute to the observed associations independently of BMI, with red and processed meat intake being an established risk factor for colorectal cancer [65], and a recent meta-analysis additionally reporting increased risk of liver, lung, breast, and endometrial cancers [9] with red and processed meat intake. Some studies have reported positive associations between red and/or processed meat intake and stomach [66], pancreatic [67], prostate [68, 69], and bladder cancer [70,71,72], and between red meat or poultry and non-Hodgkin’s lymphoma [69, 73, 74], although the evidence is not entirely consistent [9]. Higher red and processed meat intake can increase cancer risk due to higher levels of heme-iron, saturated fat and cholesterol, heterocyclic amines and polycyclic aromatic hydrocarbons formed during cooking, nitrosamines formed from nitrites and nitrates in processed meat, through adverse impacts on the microbiota, increased weight gain and metabolic perturbations, and through increased risk of various predisposing diseases (e.g. colorectal adenoma, diabetes, liver disease, gallstones, pancreatitis) [75,76,77,78,79,80]. A higher intake of fiber and whole grains [7, 81] and fruits and vegetables [8, 33, 82, 83] has also been associated with reduced risk of total, colorectal, and breast cancers, and possibly other cancers [84,85,86]. The slightly stronger association between vegan diets and reduced overall cancer risk and some specific cancers than for vegetarians may be explained by dietary differences between the groups, e.g. either higher intake of plant foods or avoidance of other animal products (such as dairy and eggs) or a combination of these. Dairy products have been associated with reduced colorectal cancer risk [87], but increased prostate cancer risk [88] and could contribute to differences between vegetarians and vegans for these two and possibly other cancers. Calcium and insulin-like growth factor-1 have been thought to contribute to the association between dairy and reduced colorectal [87], and increased prostate cancer risk [88], respectively. The lower levels of circulating IGF-1 observed in vegans [89, 90] is consistent with the positive associations observed between dairy or milk consumption and IGF-1 levels [91,92,93].
The current analysis may have some limitations that need to be discussed. Although confounding by other risk factors could be an issue, as vegetarians and vegans in general are relatively health conscious, most studies adjusted for other important risk factors such as smoking, alcohol and physical activity. In addition, similar results were observed in the Adventist Health Studies, a population with a very low prevalence of alcohol consumption and tobacco smoking [94], suggesting residual confounding by alcohol or smoking may be less likely to explain the observed associations in those studies. Among the cancers for which we observed significant associations, the estimated E-values ranged from 1.39 (lower CI: 1.11) for breast cancer to 3.04 (lower CI: 1.60) for stomach cancer, with most of the estimates around 1.8–1.9, suggesting an unadjusted confounder (or the joint impact of several unadjusted confounders) would have to be relatively strongly associated with both a vegetarian diet and cancer risk to fully explain away the observed associations. Although we cannot exclude such a possibility with complete certainty, we consider this less likely. Although we were not able to assess the quality of the vegetarian and vegan diets (e.g. if diets contained more whole plant foods or unhealthier items like refined grains, fast foods and sugary drinks), it seems likely given the results, that the majority may have had a decent diet quality, and the findings are in general consistent with reported food group intakes between diet groups [10, 95,96,97]. Heterogeneity between studies is expected as studies have been conducted in different populations with different background dietary patterns, different detail of the diet assessment tools, differences in methodological issues like adjustment for confounders and differences in the adherence to these diets over time. However, we observed little heterogeneity in the results both overall and in subgroup analyses. Publication bias can also affect meta-analyses of published studies, but we found little evidence of publication bias in these analyses. Only in the analysis of vegetarian diets and breast cancer did we find some indication of publication bias, but this seemed to be driven by one outlying study, which did not substantially influence the summary estimate when excluded. The limited number of studies limited our ability to test for publication bias and conduct subgroup and sensitivity analyses.
Measurement errors in dietary intake at baseline and changes in dietary intake during follow-up could have affected the results. Most of the studies used food frequency questionnaires to assess dietary intake. However, few studies to date have had repeated measures of dietary intake. In the EPIC-Oxford study and Oxford Vegetarian Study 73% of participants who completed a second dietary assessment 5 years after baseline remained in the same diet groups [98], while 27% changed diet group, and it is likely that a larger proportion of participants may have changed their diet with additional follow-up. A stronger inverse association was observed between a vegetarian diet and overall cancer mortality when excluding subjects who changed diet groups during follow-up instead of using only baseline data (HRs of 0.81, 0.71–0.93 vs. 0.90, 0.80–1.03) [98]. Other studies have reported stronger associations between red and processed meat consumption and total cancer mortality when using repeated vs. baseline only measurements (HRs of 1.19, 1.17, and 1.14 when using repeated measures vs. 1.08, 1.04, and 1.08 when using baseline only measures for the highest vs. lowest quintile of total red meat, unprocessed red meat, and processed red meat consumption, respectively) [99]. In the Swedish Mammography Cohort, the HRs for high red meat intake in relation to pancreatic cancer were 1.30 (0.77–2.31) when using only a baseline measurement, 1.73 (0.99–2.98) with updated averages of two measurements and 2.63 (1.02–6.84) when comparing participants with a consistently high intake with those with a consistently low intake [100]. Since all studies included in this meta-analysis only used a baseline dietary assessment to analyse the association between vegetarian and vegan diets and cancer risk it is likely that the observed associations reported here are conservative estimates of the true underlying association. Any additional studies using repeated dietary assessments and addressing the impact of long-term adherence to vegetarian and vegan diets (also throughout the follow-up) in relation to cancer incidence may be informative.
We had a limited number of studies included in most analyses and may have had insufficient statistical power to detect statistically significant associations with some cancers (e.g. distal colon, rectal, lung, endometrial, ovarian, prostate, and kidney cancers). This also precluded subgroup and sensitivity analyses as well as testing for publication bias in several analyses, and particularly in the analysis of vegans. Heterogeneity expressed by small I2-values, was generally low, however, several of the meta-analyses were based on few studies, and therefore, the ability of I2 to reliably detect heterogeneity is limited. Given the low number of studies in vegans, we also cannot exclude some degree of chance variation as the two studies reporting on vegan diets were not completely representative of the overall results for vegetarians in relation to colorectal and prostate cancer. For vegetarian diets and colorectal cancer, most studies showed associations in the direction of reduced risk, except for the combined analysis of EPIC-Oxford and Oxford Vegetarian Study, which appeared to deviate. Additional larger studies on vegans and cancer risk are needed as the current results may be considered preliminary due to the few available studies. Lastly, although we did develop a protocol for this project as part of a grant application for funding for the project, this was not registered a-priori.
Strengths of the current analysis includes the comprehensive search, analyses across a large number of cancer types, multiple subgroup and sensitivity analyses including use of E-values, and the high study quality of the included studies. These findings may have important public health implications and provide further support for the adoption of primarily plant-based diets in cancer prevention. Given the relatively low prevalence of vegetarians and vegans generally in many populations, the attributable risk could be sizeable, e.g. around 13% and 23% for total cancer, respectively if the results from the studies included are representative for the general population, however, red and processed meat consumption among the nonvegetarians in these studies is lower and intake of plant foods is higher than in the general population [10, 95,96,97], thus it is possible that these estimates could be conservative. Further and larger studies are needed to further investigate the observed associations, associations across less common and less investigated cancers and associations with vegan diets, which were investigated in a limited number of studies, and in general there were relatively few vegans included across studies. Nevertheless, the findings are largely consistent with current evidence of food groups and cancer risk, suggesting the adoption of much more plant-based diets such as vegetarian and vegan diets could play an important role in cancer prevention.
In conclusion, vegetarian diets were associated with reductions in the relative risk of total cancer (13%), stomach (43%), colorectal (14%), pancreatic (23%), breast (10%), and bladder (21%) cancers, melanoma (21%) and non-Hodgkin’s lymphoma (23%). Vegan diets were associated with a 23% reduction in risk of total cancer and a 20% reduction in breast cancer risk. Although further studies are needed to clarify the associations between vegetarian and vegan diets across less investigated cancer sites, these findings support a beneficial role of vegetarian and vegan diets in cancer prevention and provide further support for recommendations to adopt much more plant-based dietary patterns in the general population.