Observational dose response meta-analysis methods may bias risk estimates at low consumption levels: the case of meat and colorectal cancer

Background Observational studies of foods and their relation to health are known to be susceptible to bias, particularly due to confounding between diet and other lifestyle factors. Common methods for deriving dose-response meta-analysis (DRMA) relationships may contribute to biased or overly-certain estimates of risk. Objective use DRMA models to evaluate the empirical evidence for CRC association with unprocessed red meat (RM) and processed meats (PM), and the consistency of this association for low and high-consumers under different modeling assumptions. Methods Using the Global Burden of Disease project’s systematic reviews as a start, we compiled a dataset of studies of PM with 29 cohorts contributing 23,522,676 person-years and of 23 cohorts for RM totaling 17,259,839 person-years. We fitted DRMA models to lower-consumers only (consumption < US median of PM (21 g/day) or RM (56 g/day)) and compared them with DRMA models using the full consumption range. To investigate the impact of model selection we compared classic DRMA models against an empirical method for both lower consumers only, and for all consumers. Finally, we included the type of reference consumer (non-consumer or mixed consumer/non-consumer) as a covariate in a multivariate meta-analysis of the lowest consumption arm. Results We found no significant positive association for RM (RR at 50 g/day (1.04 (0.99-1.10)) or PM (RR at 20g/day 1.01 (0.87-1.18)) with CRC based on any DRMA model type when using only lower consumers. Only the full range of consumption yielded association with CRC, and the empirical DR showed non-linear, non-monotonic relationships. We did not find significant able. Conclusions These results show risk overestimation at low consumption can result from modeling assumptions and from the influence of higher consumption amounts. Furthermore, our results show that a no-risk limit of 0 g/day consumption of RM and PM is inconsistent with the evidence. Statement of Significance This article describes critical issues in classical methods of dose-response modeling that may introduce and exacerbate bias, leading to overestimates of risk at low levels of consumption. We propose alternative methods which quantitatively reflect the uncertainty in dose-response meta-analytic models and show that risk overestimation at low consumption can result from modeling assumptions and from the influence of higher consumption amounts. The example case of unprocessed and processed meat and colorectal cancer is used to demonstrate methods for dose-response that can be adapted to other observational evidence of dose-dependent risk and could be used in developing dietary guidelines in a transparent and systematic way.