Patterns of Red and Processed Meat Consumption and Risk for Cardiometabolic and Cancer Outcomes

Observational studies have reported higher incidence of all-cause mortality, cardiometabolic diseases, and cancer outcomes in people who consume greater quantities of red meat (1–6). Consequently, most guidelines from national and international agencies recommend limiting intake of red and processed meat (7–9). However, additional scrutiny of the evidence to determine the extent to which current recommendations are justified is warranted, particularly given the possible methodological limitations of systematic reviews to date (for example, the lack of consideration of the overall certainty of evidence) and other confounding factors (10, 11).

Foods and nutrients are not consumed in isolation, and their effects may differ depending on the totality of one's diet and how dietary habits change over time. Moreover, interventions focusing on modification of intake of particular foods or nutrients require compensatory changes in other dietary components. Nevertheless, most nutritional epidemiologic research since the 1970s has focused on the effects of individual foods or nutrients (12). Given the potential for interaction, an increasingly common alternative to focusing on individual foods or nutrients is to examine the effects of dietary patterns on health outcomes (13). Two approaches are commonly used to define dietary patterns: data-driven methods, including factor analysis or principal-components analysis, or a priori approaches that use diet indices or scores based on dietary recommendations or characteristics.

This review was done to inform recommendations on red and processed meat intake from the NutriRECS (Nutritional Recommendations) consortium (14). We conducted 4 additional systematic reviews addressing evidence from randomized trials on the effect of red meat consumption on health outcomes (15), observational evidence on the association between red and processed meat consumption and cardiometabolic outcomes (16), observational evidence on the association between red and processed meat consumption and cancer outcomes (17), and qualitative and quantitative evidence on public values and preferences regarding meat consumption (18). We used the results of these to develop guideline recommendations on red and processed meat consumption (19). In this article, we report the results of a systematic review addressing the association between dietary patterns that are lower versus higher in red and processed meat intake and the risk for cardiometabolic and cancer outcomes.

Methods

This article complies with the recommendations of PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) (20). We registered the protocol in PROSPERO (CRD42017074074) on 10 August 2017.

Data Sources and Searches

With assistance from an experienced librarian, we developed a comprehensive search strategy for 5 databases: MEDLINE, EMBASE, the Cochrane Central Register of Controlled Trials, Web of Science, and CINAHL. We searched each database without restrictions on year or language of publication from inception to 8 July 2018, with an updated search of MEDLINE through to April 2019 (Supplement Table 1). In addition, we searched the following gray literature sources: ProQuest Dissertations & Theses Global (1989 to 2018), trial registries (ClinicalTrials.gov and the World Health Organization International Clinical Trials Registry Platform Search Portal), bibliographies of included articles, and relevant literature reviews. This search strategy informed all supporting NutriRECS reviews on red and processed meat (15–17) except the one addressing public values and preferences (18).

Study Selection

We included cohort studies with 1000 or more participants that reported an association between dietary patterns and 1 or more of our outcomes of interest in adults with or without cardiometabolic conditions but without cancer or any infectious or chronic noncardiometabolic conditions. We excluded studies that did not report the quantity of consumption of red and processed meat across categories of dietary habits. Red meat was defined as mammalian meat, and processed meat was defined as white or red meat preserved by smoking, curing, salting, or adding preservatives (21). We assumed serving sizes of 120 g for unprocessed red meat, 50 g for processed meat, and 100 g for mixed unprocessed red and processed meat. These were selected to be comparable to serving sizes used in other systematic reviews and to reflect those used by the U.S. Department of Agriculture and the U.K. Food Standards Agency (4–7). We also included studies comparing vegetarians with nonvegetarians. When more than 1 eligible article reported on the same exposure and cohort and addressed the same outcome, we included only results from the article with the longest follow-up. If the duration of follow-up was the same across articles, we included the article with the largest number of participants, resulting in each unique cohort study as the unit of analysis.

The panel for the NutriRECS guideline on red and processed meat, which comprised members of the public and clinicians, including dietitians, epidemiologists, and methodologists, selected the outcomes of interest for this systematic review (14, 22). These included major cardiometabolic morbidity and mortality; incidence of or mortality associated with gastrointestinal, breast, gynecologic, and prostate cancer; quality of life; and satisfaction with diet.

Reviewers conducted pilot screening exercises and received detailed instructions for each item before screening. Pairs of reviewers independently screened titles and abstracts in duplicate and reviewed the full-text articles of those found to be potentially eligible. Reviewers resolved disagreements by discussion or, if necessary, by third-party adjudication.

Data Extraction and Risk-of-Bias Assessment

Reviewers conducted calibration exercises and worked in duplicate to independently extract data. We used a standardized, piloted data abstraction form. The reviewers resolved disagreements by discussion or, if necessary, by third-party adjudication involving a senior investigator. We extracted the following information from each study: setting, number of participants at baseline and follow-up, age, sex, method of diet assessment, dietary pattern data (intake of red and processed meat), type of cancer or cardiometabolic disease, years of follow-up, and effect estimates and corresponding 95% CIs.

Reviewers, independently and in duplicate, assessed risk of bias of each eligible study by using a modified version of the CLARITY (Clinical Advances Through Research and Information Translation) instrument (23, 24), with omission of 1 item related to co-interventions that was not relevant to our review (Supplement Table 3). We rated each item as having definitely low, probably low, probably high, or definitely high risk of bias. We adapted item 2 (assessment of exposure) to address the validity of dietary measures. For example, if a study measured diet at least once every 5 years using a food-frequency questionnaire validated against a weighted food record for red and processed meat, it was deemed to have definitely low risk of bias for this item. We also adapted item 4 (adjustment for prognostic factors) on the basis of established prognostic factors for each outcome of interest. Consultation with research methodologists and nutrition researchers confirmed the appropriateness of the instrument and informed criteria to evaluate each item. We considered each item to be equally important. Risk of bias was evaluated at the outcome level and not the study level. We considered studies that were rated to have high risk of bias on 2 or more of the 7 domains to have high overall risk of bias. Although this threshold is arbitrary, it represents a compromise between excessive stringency and leniency and facilitates subgroup analysis by providing an approximately even division between studies at higher versus lower risk of bias.

Data Synthesis and Analysis

The lowest category (for example, tertiles or quartiles) of adherence to dietary patterns high in red and processed meat was compared with the highest category as a proxy to determine the dietary patterns lower versus higher in red and processed meat consumption. When a single study investigated multiple dietary patterns, our analysis focused on the one with the greatest difference in red or processed meat intake between the lowest and highest categories. In studies that investigated multiple dietary patterns, we used factor analysis or principal-component analysis to analyze the patterns with the highest factor loadings for red meat, processed meat, or both. Using the Hartung–Knapp–Sidik–Jonkman model (25, 26) and a DerSimonian–Laird random-effects model as a sensitivity analysis (27), we conducted meta-analyses comparing the lowest and highest categories of intake. For studies in which the outcomes of interest were reported stratified by participant characteristics (for example, sex) with no overlap in participants or events across strata, we meta-analyzed across subgroups. We calculated the risk difference by multiplying the pooled effect estimate from our meta-analyses by the baseline risk for cancer incidence or mortality, which was based on GLOBOCAN cumulative lifetime risks (28, 29). For cardiometabolic disease, data from the Emerging Risk Factors Collaboration (30) provided the baseline risk to calculate risk differences over 10.8 years.

We conducted subgroup analyses for each outcome based on our assessment of overall risk of bias for each study. When we found a statistically significant subgroup effect based on risk of bias, we present results for studies with low risk of bias. We also conducted a subgroup analysis based on the methods for defining dietary patterns.

Certainty of Evidence

One investigator assessed the certainty of evidence as high, moderate, low, or very low for each outcome using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) criteria (31), with all assessments confirmed by senior investigators. According to the GRADE approach, observational studies may provide moderate- or high-certainty evidence if they show a large magnitude of effect or a dose-response gradient and when suspected biases work against the observed direction of effect. Observational studies without these characteristics provide low-certainty evidence, and those that are also limited by risk of bias, inconsistency, indirectness, imprecision, or publication bias provide very-low-certainty evidence (32–37).

Role of the Funding Source

This study received no funding.

Results

Of the 13 154 unique citations initially identified by our search, 11 653 were deemed ineligible after title and abstract screening (Figure). Among the 1501 full-text articles evaluated, 105 were potentially eligible, yielding 70 unique cohorts with 6 035 051 participants. Most of the cohorts (83%) were also included in the accompanying systematic reviews (16, 17) to support the guideline recommendation.

• Download figure
• Download PowerPoint

The articles were published between 1994 and 2018 and included 1804 to 492 382 participants followed for 2.0 to 34.3 years. The mean age of participants at enrollment ranged from 33 to 71 years, with most studies including a majority of participants aged 50 to 60 years, and the proportion of women ranged from 0% to 100%, with most studies including a similar number of men and women. Six studies recruited participants with preexisting cardiometabolic conditions, the most common being hypertension. Supplement Table 2 shows additional study characteristics.

Investigators used a wide variety of methods to define dietary patterns (Supplement Table 2). In 63 cohort studies (60%), a posteriori dietary patterns derived from factor analyses, principal-component analyses, cluster analyses, or reduced-rank regressions were used. Among studies using factor analysis or principal-component analysis, dietary patterns were derived by the covariance matrix of individual foods to reduce the dimensionality from a high number of foods to a few patterns of food consumption that explained the maximum variation in dietary habits (13). The resultant patterns were aggregates of foods that were highly correlated with one another. Patterns derived by principal-component analysis were linear combinations of the observed variables, whereas factors derived by factor analysis were latent constructs. The emerging patterns were often adjusted using an “orthogonal rotation” so that the final patterns were uncorrelated. For each pattern, summary factor scores were obtained that defined each participant's degree of adherence. Among studies in which reduced-rank regression was used, linear combinations of foods that maximally explained a set of intermediate or surrogate measures of the outcome of interest (such as biomarker levels) were derived and pattern scores were obtained that, similar to factor scores, represented each participant's degree of adherence to the pattern (38). In studies that used cluster analysis, the k-means method was used to identify aggregates of participants with similar dietary habits (13).

In 15 studies, dietary patterns were defined a priori using indices or scoring systems. Of the eligible studies, 19 investigated vegetarian versus nonvegetarian participants for at least 1 of our outcomes of interest. Our subgroup analyses found no differences among the various methods used to determine dietary patterns for any outcome (Supplement Table 4).

Red and processed meat intake varied widely across categories. Thirty studies recorded a quantitative estimate of red or processed meat consumption (for example, grams per day or servings per week), and others solely reported the factor loadings or compared vegetarians versus nonvegetarians without information on meat consumption. Among the 27 studies reporting on red meat intake (unprocessed, unspecified, or mixed), the difference between extreme adherence categories was less than 2 servings per week in 6 studies, 2 to 5 servings per week in 17 studies, and more than 5 servings per week in 4 studies. In the 19 studies reporting on intake of processed meat, the difference between extreme adherence categories was less than 2 servings per week in 4 studies, 2 to 5 servings per week in 13 studies, and more than 5 servings per week in 2 studies.

Risk-of-Bias Assessment

Supplement Table 3 presents risk of bias for each study and outcome. The percentage of studies with high overall risk of bias varied across outcomes (10 of 24 [42%] for all-cause mortality, 13 of 25 [52%] for cardiovascular mortality, 3 of 5 [60%] for fatal and nonfatal stroke, 0 of 2 [0%] for myocardial infarction, 7 of 12 [58%] for cardiovascular disease, 2 of 3 [67%] for overall cancer incidence, 8 of 18 [44%] for overall cancer mortality, and 9 of 14 [64%] for type 2 diabetes). The most common methodological limitations were lack of repeated measurement of intake in the dietary patterns, use of a measure that was not validated for red and/or processed meat, and inadequate adjustment for potential confounders. We did not find significant differences in any outcome for the studies judged to have high versus low risk of bias (Supplement Table 5).

All-Cause Mortality

We found a small decrease in risk for all-cause mortality associated with dietary patterns lower in red or processed meat intake. Evidence was rated to have very low certainty due to inconsistency (Table 1).

Table 1. Summary of Findings for Lower Adherence to Dietary Patterns High in Red and Processed Meat Intake and Risk for Cardiometabolic Outcomes

Cardiovascular Outcomes

Dietary patterns lower in red and processed meat intake were associated with decreased risk for cardiovascular mortality, based on very-low-certainty evidence due to inconsistency (Table 1). Low-certainty evidence showed a small reduction in risk for nonfatal stroke for dietary patterns low in red and processed meat intake but no statistically significant associations for risk for overall stroke and fatal stroke.

We did not observe a statistically significant association between dietary patterns and risk for fatal and nonfatal myocardial infarction, fatal and nonfatal cardiovascular disease, and nonfatal cardiovascular disease (Table 1).

Type 2 Diabetes

We found a very small reduction in risk for type 2 diabetes associated with dietary patterns lower in red or processed meat consumption (Table 1). The overall certainty of this evidence was very low due to inconsistency.

Cancer Outcomes

Dietary patterns lower in red and processed meat intake were associated with a small to very small reduction in risk for overall cancer incidence and mortality (Table 2). However, this evidence was considered to have very low certainty for overall cancer incidence (due to imprecision) and overall cancer mortality (due to inconsistency). No statistically significant risk was found for dietary patterns lower in red and processed meat intake for incidence of breast, colorectal, endometrial, liver, ovarian, pancreatic, prostate, stomach, and uterine cancer. Similarly, we found no differences among dietary patterns in risk estimates for mortality associated with breast, colorectal, esophageal, liver, ovarian, prostate, and stomach cancer.

Table 2. Summary of Findings for Lower Adherence to Dietary Patterns High in Red and Processed Meat Intake and Risk for Cancer Incidence and Mortality

We found low-certainty evidence that dietary patterns lower in red or processed meat consumption were associated with a very small reduction in risk for incidence of extrahepatic and gallbladder cancer. Low-certainty evidence also suggested that risk for death due to pancreatic cancer was lower for the dietary patterns with low intake of red or processed meat.

Discussion

In this systematic review and meta-analysis of 70 unique cohorts with 6 035 051 participants, we found low- to very-low-certainty evidence that dietary patterns lower in red or processed meat intake result in a small to very small reduction in risk for all-cause mortality, cardiovascular mortality, fatal coronary heart disease, fatal myocardial infarction, overall cancer mortality, overall cancer incidence, and type 2 diabetes (Tables 1 and 2).

Strengths of our review include a priori–defined methods (22); a comprehensive search of 6 primary databases using a strategy developed by an expert librarian; duplicate screening of studies for eligibility, data extraction, and assessment of risk of bias; and inclusion of a large number of studies and participants. Further, we used the GRADE approach to assess the certainty of evidence and to present absolute estimates of effect for all 30 outcomes.

One of the primary limitations of our work is the heterogeneity of dietary patterns across studies. Although all patterns discriminated between participants with low and high intake of red and processed meat, other food and nutrient characteristics of dietary patterns and the quantity of red and processed meat consumed varied widely across studies. Moreover, the quantity of red and processed meat consumed differed across dietary patterns and studies. For example, one study compared 1.4 versus 3.5 servings of processed meat per week (39), whereas another compared 0.7 versus 4.9 servings per week (40). Such inconsistencies may have increased heterogeneity of meta-analyses and potentially reduced the magnitude of observed associations. Also, analyses of extreme categories of adherence may artificially inflate effect estimates and may not be indicative of effects observed at typical levels of adherence. Second, we were unable to analyze the data separately for red and processed meat because authors typically combined them or did not distinguish between them in primary studies. Third, all eligible studies used observational designs and were thus prone to confounding. Although we minimized confounding by using the most adjusted analyses from each study in our meta-analyses, residual confounding remains a plausible explanation for all associations. Finally, eligible studies used recall-based methods for dietary measurement that are prone to measurement error, which can result in either an underestimate or an overestimate of observed associations (41, 42).

Although previous reviews have reported a positive association between red and processed meat intake and all-cause mortality, cardiovascular disease, stroke, myocardial infarction, and type 2 diabetes (4, 6, 10, 11), our systematic review is, to our knowledge, the first to address dietary patterns with respect to red and processed meat consumption and to include an assessment of the certainty of evidence. The 2010 Dietary Guidelines Advisory Committee of the U.S. Department of Agriculture conducted a project that summarized the evidence on dietary patterns and their effect on health, including obesity, cardiovascular disease, and type 2 diabetes (43). In general, although some dietary patterns, such as low-carbohydrate diets, were not considered, no single dietary pattern was associated with more favorable health outcomes. However, the Mediterranean-style diet, the DASH (Dietary Approaches to Stop Hypertension) diet, and Dietary Guidelines–related patterns were consistently associated with positive health outcomes (43).

Our work complements other systematic reviews performed to address NutriRECS recommendations on red and processed meat consumption. Although we found a small to very small decrease in risk for several adverse health outcomes, the effect sizes observed in our dose-response meta-analyses that directly addressed red and processed meat were, in general, only slightly smaller (16, 17). For example, studies directly addressing meat consumption suggested that a reduction of 3 servings per week resulted in 8 (red meat) and 9 (processed meat) fewer deaths per 1000 persons (16), whereas we found 15 fewer deaths per 1000 persons among those adhering to dietary patterns lower in red and processed meat consumption. Similarly, studies directly addressing meat consumption found that a reduction of 3 servings per week may decrease lifetime risk for cancer death by 7 (red meat) and 8 (processed meat) deaths per 1000 persons (17) versus 12 deaths per 1000 persons observed in this review. If red and processed meat were causally driving the association between diet and adverse health outcomes, we would anticipate finding stronger associations in our systematic reviews specific to red and processed meat intake (14, 22), but we did not. Our findings indicate the possibility that dietary components associated with red and processed meat intake may confound its association with health outcomes. However, inferences about causality should be interpreted with caution given the low- to very-low-certainty evidence.

Because of concerns about adverse health outcomes, recent dietary guidelines recommend limiting red and processed meat intake in cultures with traditionally high red meat consumption (1, 7–9). The results of our systematic review raise concerns about these recommendations. First, we identified only low-certainty (and often very-low-certainty) evidence linking dietary patterns lower in red and processed meat intake with small reductions in adverse cardiometabolic and cancer outcomes, making causal inferences tenuous. Second, our results suggest that dietary patterns lower in red or processed meat intake are associated with a small or, in most cases, very small reduction in risk for all-cause mortality, 3 cardiometabolic outcomes, and 5 cancer outcomes. Third, our findings of stronger associations in studies of dietary patterns high in red and processed meat intake compared with studies directly addressing red and processed meat consumption suggest the possibility of confounding by other dietary characteristics correlated with red and processed meat consumption.

In conclusion, adherence to dietary patterns lower in red or processed meat intake may result in decreased risk for all-cause mortality, cardiometabolic disease and mortality, and cancer morbidity and mortality. Nevertheless, the magnitude of these effects for all outcomes is small to very small, and the certainty of evidence is low to very low. Our results therefore raise questions about the plausibility of red and processed meat being causally related to adverse health outcomes.