Cancer is a leading cause of death globally. The World Cancer Research Fund (WCRF) identifies diet and nutrition as modifiable risk factors. However, existing meta-analyses of observational studies on diet and cancer risk may be influenced by biases such as measurement error, confounding, and publication bias. This umbrella review systematically evaluates the robustness of observational meta-analytic evidence for associations between food/nutrient intake and cancer risk at 11 anatomical sites. The aim is to assess the strength and validity of this evidence and determine the potential value of further research in this area. The inherent challenges in nutritional epidemiology, specifically the reliance on self-reported dietary data and the complexities of converting food intake into nutrient intake using food composition databases, underscore the need for a critical evaluation of the existing evidence base.

The researchers reviewed 860 meta-analytic comparisons from the WCRF Third Expert Report, focusing on the association between various dietary factors and cancer risk across 11 anatomical sites. The review included data from prospective cohort studies and, in the case of head and neck cancer, case-control studies. The meta-analyses employed both continuous and dichotomous exposure contrasts. Existing research was evaluated for heterogeneity, small study effects, and excess significance bias, and the certainty of evidence was graded using established criteria. The review also incorporated fail-safe numbers and conditional power analysis to assess the likelihood that future research would alter the conclusions drawn from the existing meta-analyses.

The umbrella review included 860 meta-analytic comparisons from the WCRF Third Expert Report, encompassing data on 11 cancer sites. Data were extracted at both the meta-analysis and individual study levels, including dietary factors, comparison contrasts, cancer outcomes, study characteristics, and effect estimates. A random-effects model was used to estimate summary associations and their confidence intervals. Heterogeneity was assessed using the I² statistic, and 95% prediction intervals were calculated to account for heterogeneity. Egger's test was used to detect small study effects, and an algorithm based on non-central t-distribution was applied to assess excess significance bias. Meta-analyses were graded into four evidence categories (strong, highly suggestive, suggestive, weak) based on multiple criteria including statistical significance, sample size, heterogeneity, prediction intervals, and bias assessment. Fail-safe numbers (FSN) and conditional power (CP) analyses were conducted to determine the potential impact of future research on the existing evidence.

Of the 860 meta-analyses, only 10 (1.2%) were supported by strong evidence, and 13 (1.5%) by highly suggestive evidence. Strong evidence primarily involved alcohol's positive association with colorectal (CRC) and breast cancer risk, and inverse associations between dairy/calcium/whole grains and CRC risk. Highly suggestive evidence largely centered on alcohol's positive association with various cancers (colon, rectum, esophagus, head and neck, liver) and coffee's inverse association with liver cancer and skin basal cell carcinoma. Suggestive evidence (4.9%) included associations between red/processed meat and CRC risk, inverse associations between fiber and breast cancer risk, and other associations. A substantial number (21%) of meta-analyses had weak evidence, and the majority (71%) had non-significant findings. Analyses of FSN and CP suggested that further research is unlikely to substantially alter the current conclusions, with few exceptions mainly involving null associations for understudied cancers and dietary factors.

This umbrella review provides a comprehensive evaluation of the evidence linking diet and cancer risk. The findings generally support the limited number of robust associations previously identified, validating the cautious approach in interpreting observational studies' findings. The study's rigorous methodology, including multiple bias assessments and research synthesis metrics, enhances the reliability and generalizability of the results. The limited number of strong and highly suggestive associations highlights the complexity of this research area and the need for improved research methods.

This comprehensive review found strong or highly suggestive evidence for a few key associations between diet and cancer risk, notably the positive association between alcohol and several cancers and the inverse association between certain foods (dairy, whole grains) and colorectal cancer. Other associations warrant further investigation. Future research should prioritize the development of more robust dietary assessment methods, exploration of overall dietary patterns rather than single nutrients/foods, and investigation into the complex biological mechanisms that underlie diet-cancer relationships. Public health efforts should concentrate on addressing known major diet-related risk factors like obesity and alcohol consumption.

The review relied on observational studies, limiting the ability to infer causality. Dietary assessment methods (primarily food frequency questionnaires) are prone to measurement error, potentially attenuating risk estimates. The statistical tests used for bias assessment do not definitively prove the absence of bias. The review might have missed some studies not included in the WCRF data, and the analyses were limited to those reported in the original meta-analyses. Finally, the selection of specific criteria for grading evidence involved some degree of arbitrariness.