Red and Processed Meat Consumption and Risk for All-Cause Mortality and Cardiometabolic Outcomes

Abstract

Growing evidence shows an increased risk for cardiometabolic disease associated with the consumption of red and processed meat. Although previous systematic reviews reported positive associations between red meat intake and all-cause mortality (1), cardiovascular mortality (2), and stroke (3) and between processed meat consumption and all-cause mortality (1, 4), cardiovascular mortality (2), stroke (3), coronary heart disease (5), and type 2 diabetes (5), results have not been consistent. One review did not find an association between unprocessed red meat and all-cause mortality (4), and another found no association with cardiovascular disease (5). Although Aune and colleagues (6) reported a relationship between red meat intake and type 2 diabetes, Micha and colleagues (5) did not detect this association in a review published 1 year later. Methodological limitations in previous reviews included failure to address risk of bias of primary studies (for example, references 3 and 6), lack of evaluation of certainty of evidence (for example, references 2 to 6), and failure to consider the magnitude of observed effect (for example, references 2 to 6). These limitations may have affected the credibility of recommendations issued by governments and authoritative organizations regarding red and processed meats. As part of NutriRECS (Nutritional Recommendations and accessible Evidence summaries Composed of Systematic reviews), a new initiative to establish trustworthy dietary recommendations that meet internationally accepted standards for guideline development, we developed guidelines addressing red and processed meat consumption (7). To inform these recommendations, we conducted 5 systematic reviews of the evidence (811). Here, we present results from a systematic review of cohort studies addressing the association between red and processed meat consumption and all-cause mortality, cardiometabolic outcomes, quality of life, and satisfaction with diet among adults. Methods We registered a protocol for this review at PROSPERO (CRD42017074074) in August 2017. Data Sources and Search Strategy An experienced research librarian developed the search strategy, which was used across all supporting reviews except the one addressing public values and preferences (Supplement 1). We searched MEDLINE, EMBASE (Elsevier), Cochrane Central Register of Controlled Trials (Wiley), Web of Science (Clarivate Analytics), CINAHL (EBSCO), and ProQuest from inception. We also reviewed reference lists of relevant systematic reviews. The final search of all databases included references up to July 2018, except for the MEDLINE search, which included references up to April 2019. Supplement. Search Strategy and Supplement Tables (Supplement 1) and Figures and Technical Appendix (Supplement 2) Study Selection We included cohort studies in any language that enrolled at least 1000 adults, compared participants consuming different amounts of unprocessed red meat or processed meat, and reported on 1 or more of our outcomes of interest. Red meat and processed meat were defined, respectively, as mammalian meat and white or red meat preserved by smoking, curing, salting, or adding chemical compounds (for example, hot dogs, charcuterie, sausage, ham, and deli meats) (12). We also included studies comparing vegetarians with nonvegetarians for sensitivity analyses. Our outcomes of interest were determined in consultation with our guideline panelwhich comprised members of the public, clinicians, epidemiologists, and methodologistsand include all-cause mortality, cardiovascular mortality (or fatal coronary heart disease or fatal myocardial infarction [MI]), cardiovascular disease (or coronary heart disease), stroke, MI, type 2 diabetes, anemia, quality of life, and satisfaction with diet. For studies reporting on ischemic and hemorrhagic stroke separately, we included results only for ischemic stroke in our meta-analyses (13). Cohorts in which more than 20% of the sample was younger than 18 years, had a noncardiometabolic disease (such as cancer), or was pregnant at baseline were excluded. We also excluded studies in which diet was assessed before adulthood, participants were asked to recall their diet before adulthood, or dietary assessments were completed by proxies, as well as studies that reported on specific components of red meat (such as iron or fat) or specific types of red meat (such as lamb). However, we did include studies reporting on beefpork combinations because beef and pork account for most red meat intake in most Western populations (14, 15). If we encountered more than 1 eligible article on the same exposure and cohort and addressing the same outcome, we included results only from the study with the longest follow-up. If the follow-up was the same, we chose the study with the most participants. Pairs of reviewers completed calibration exercises, after which they performed screening independently and in duplicate, with disagreements resolved by discussion or through third-party adjudication by an expert research methodologist. Screening was done in 2 stages: First, the reviewers assessed titles and abstracts; then, for those deemed potentially eligible, they evaluated the full-text articles. Data Extraction and Quality Assessment Using standardized, pilot-tested forms, reviewers completed calibration exercises and worked in pairs to independently extract the following information from eligible studies: cohort characteristics (such as cohort name and country), participant characteristics (including age and proportion who were female), diet characteristics (such as frequency and quantity of consumption of unprocessed red meat or processed meat), and outcomes (including absolute and relative effect measures for outcomes of interest and measures of variability). Disagreements between pairs of extractors were resolved through discussion or by third-party adjudication by an expert research methodologist. Reviewers, working independently and in duplicate, assessed each study s risk of bias by using the CLARITY (Clinical Advances Through Research and Information Translation) risk-of-bias instrument for cohort studies, omitting an item related to co-interventions that was not relevant to our review (16). Disagreements were resolved through discussion or by third-party adjudication. Research methodologists and nutrition researchers were consulted to confirm the appropriateness of the CLARITY instrument and to advise us regarding criteria for evaluating each of its items. The instrument and detailed guidance are presented in Supplement Table 1. Studies rated as high risk of bias on 2 or more of the 7 domains were considered to have a high overall risk of bias. This threshold, although somewhat arbitrary, represents a compromise between excessive stringency and leniency. Data Synthesis and Analysis We conducted separate analyses for unprocessed red meat, processed meat, and mixed unprocessed red and processed meat. If an article reported on red meat and did not specify whether it was processed or unprocessed, we assumed that it included both unprocessed and processed red meat. We included such studies in the analysis of mixed unprocessed red and processed meat because most processed meat is typically consumed as red meat (17, 18). For our primary analyses, we conducted a random-effects doseresponse meta-analysis using methods proposed by Greenland and Longnecker (19) and Orsini and colleagues (20). These methods require knowledge of the distribution of events and number of participants or person-years and mean or median quantity of intake across categories of exposure. When results from studies were analyzed across quantiles of intake but person-years or number of participants was not reported within each quantile, we estimated these values by using figures reported for the total population and dividing the total person-years or total number of participants by the number of quantiles. For studies reporting effect estimates stratified by participant characteristics (such as sex), we meta-analyzed across subgroups by using the fixed-effects model. For studies that treated the exposure as a continuous predictor in a logistic regression and did not present categorical analyses, we calculated a regression coefficient based on the relative effect reported and meta-analyzed these regression coefficients with effects from other studies obtained via the estimation method described by Greenland and Longnecker (19). These studies were excluded from the nonlinear analyses. For analyses including 5 or more studies, we tested for nonlinearity by using restricted cubic splines with knots at 10%, 50%, and 90% and a Wald-type test. For analyses in which we observed statistically significant nonlinear associations, we present results from the nonlinear model. For studies reporting the intake of red meat or processed meat as a range of values, we assigned the midpoint of upper and lower boundaries in each category as the average intake. If the highest or lowest category was open ended, we assumed that the open-ended interval was the same size as the adjacent interval. For studies reporting exposure as number of servings, we assumed that each serving of unprocessed red meat was equal to 120 g; processed meat, 50 g; and mixed unprocessed red and processed meat, 100 g. These serving sizes were selected for comparability with those used in other systematic reviews, as well as to reflect serving sizes used by the U.S. Department of Agriculture and United Kingdom Food Agency (13, 2125). We report results corresponding to the effects of a reduction in unprocessed red or processed meat intake of 3 servings per week. We used the dosresmeta package in R, version 3.5.1 (R Foundation for Statistical Computing), for our doseresponse meta-analyses (26). Further details about these meta-analyses, including sample code, are presented in Supplement 2. As a secondary analysis, we used the Har