Introduction

Growing uncertainty and changing views on the role of butter in cardiovascular disease (CVD) have been prominently discussed, including in the New York Times and Time Magazine. [1, 2] This has partly arisen from increasing controversy on the utility of focusing on isolated macronutrients, such as saturated fat, for determining risk of chronic diseases. Mounting evidence indicates a need to shift away from isolated macronutrients toward food-based paradigms for investigating dietary priorities for chronic diseases. [3, 4] The 2015 Dietary Guidelines Advisory Committee (DGAC) recommended replacing animal fats, including butter, with non-hydrogenated vegetable oils high in unsaturated fats and relatively low in saturated fatty acids. [4] Yet, the DGAC also concluded that further research was needed on the effects of saturated fat from different food sources, including animal products, on cardiovascular risk, because different food sources contain varying specific fatty acid profiles as well as other constituents that may result in distinct lipid and metabolic effects. [4]

For example, growing evidence supports potential metabolic benefits of certain dairy products, such as yogurt and possibly cheese, on risk of type 2 diabetes [5, 6], which may even relate to benefits of dairy fat. [7–9] However, the relationship of butter, which is highest in dairy fat, with diabetes remains unclear. The long-term effects of butter consumption on other major endpoints, such as all-cause mortality and CVD, are also not well-established. Previous reviews have evaluated only some of these outcomes, included butter as part of a wider investigation into dairy foods or types of fats [10–12], and utilized methods that provided imprecise estimates of effect, precluded dose-response evaluation, or may have introduced unintended bias (e.g., due to inclusion of crude, unadjusted effect estimates).

A systematic review of the evidence for of the relationship between butter consumption and long-term health is of considerable importance, both for understanding food-based health as well as informing dietary recommendations for clinicians and policy makers. The US Department of Agriculture has documented a 40-year record high in US butter consumption in 2014 [13], making a synthesis of the evidence on butter and major chronic diseases highly relevant and timely.

To synthesize the evidence on the long-term association of butter consumption with major health endpoints, we conducted a systematic review and meta-analysis of prospective observational studies or randomized clinical trials investigating butter consumption and all-cause mortality, CVD including coronary heart disease (CHD) and stroke, and type 2 diabetes in general populations.

Materials and Methods

We followed Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines (S1 File) for observational studies and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (S2 File) guidelines for trials during all stages of design, implementation, and reporting.

Results

Study characteristics

Among 5,770 unique abstracts, we identified 9 publications including 15 country-specific cohorts (Fig 1), together reporting on over 636,000 unique participants with 6.5 million person-years of follow-up and including 28,271 total deaths, 9,783 cases of incident CVD, and 23,954 cases of incident diabetes (Table 1). No randomized clinical trials evaluating butter and these endpoints were identified. Outcomes in each study were generally assessed by review of medical records, linkage to death certificates, or hospital registers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of study identification and selection process.

All systematic review and meta-analyses methods conducted according to PRISMA guidelines.

https://doi.org/10.1371/journal.pone.0158118.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of 9 prospective studies providing risk estimates for the association between butter intake and total mortality, CVD and type 2 diabetes.

https://doi.org/10.1371/journal.pone.0158118.t001

Diet was generally assessed by detailed, semi-quantitative food frequency questionnaires; one cohort utilized a structured diet history interview (Table 1). The median butter consumption across studies ranged from 4.5g/d (0.3 servings/d) in the European Prospective Investigation into Cancer and Nutrition (EPIC) studies to 46 g/d (3.2 servings/d) in Finland. Mean participant age ranged from 44 to 71 years. All studies were published between 2005 and 2015, and included 1 in the Netherlands, 2 in the US, 2 in Finland, 2 in Sweden, and 2 from the multi-country, multi-cohort EPIC study which included 8 country-specific cohorts from Denmark, France, Italy, Germany, the Netherlands, Spain, Sweden and the UK. Five of the studies presented results from models with optimal covariate adjustment including demographics, clinical risk factors, and other dietary habits; the remainder provided results with moderate covariate adjustment.

All-cause mortality

Two large studies including 9 country-specific cohorts evaluated butter intake and all-cause mortality, including 379,763 participants and a total of 28,271 deaths. [27, 30] Pooling these studies, each daily serving of butter (14g/d) was associated with a 1% higher risk of death (RR = 1.0134 (95%CI = 1.0003, 1.0266; P = 0.045) (Fig 2). No significant heterogeneity was identified (I² = 0.0%). Findings were similar when explored using fixed-effects (P = 0.045).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Butter consumption and risk of all-cause mortality.

Within-study dose-response RRs were derived from reported linear effects or generalized least-squares trend estimation for studies reporting categories of intake, and pooled using both inverse-variance weighted random and fixed effects meta-analysis.

https://doi.org/10.1371/journal.pone.0158118.g002

CVD

Butter consumption and incident CVD (total CVD, CHD, or stroke) was investigated in 5 studies from 4 cohorts, including 175,612 participants and 9,783 cases of any CVD. When pooled, butter intake was not significantly associated with CVD (RR = 1.00 (95%CI = 0.98, 1.02; P = 0.704) (Fig 3). Results were similar with fixed effects; with minimal heterogeneity between studies (I² = 0.0%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Butter consumption and risk of any and total cardiovascular disease, stroke only and CHD only.

Data from 4 prospective cohorts with 175,612 participants and 9,783 cases for CVD, 3 cohorts of 173,853 participants and 5,299 events for stroke, and 3 studies of 149,056 participants and 4,484 cases of CHD. Within-study dose-response RRs were derived from reported linear effects or generalized least-squares trend estimation for studies reporting categories of intake, a pooled using both inverse-variance weighted random and fixed effects meta-analysis. CHD: Coronary Heart Disease; CVD: Cardiovascular Disease; D+L: DerSimonian and Laird random effects; I-V: Inverse-variance fixed effects; RR (95%CI): Relative Risk and 95% Confidence Interval.

https://doi.org/10.1371/journal.pone.0158118.g003

Stroke alone was investigated in 3 studies including 5,299 incident cases (Fig 3); and CHD alone, in 3 studies including 4,484 cases. No significant associations were seen: RR = 1.01 (95%CI = 0.98, 1.03; P = 0.737), and RR = 0.99 (95%CI = 0.96, 1.03; P = 0.537), respectively, without evidence for heterogeneity (I² = 0.0% each).

Total CVD, combining outcomes of CHD and stroke was reported in two cohorts, consisting of 123,497 participants, with 6051 total CVD events (4033 CHD and 2018 stroke). Results from random and fixed effects meta-analysis were identical (RR = 0.99 (95%CI = 0.96, 1.02); P = 0.498). No heterogeneity between the two studies was detected (I² = 0.0%, P = 0.498).

Type 2 diabetes

Four studies including 11 country-specific cohorts reported on butter consumption and onset of type 2 diabetes [14, 15, 26, 29], including 201,628 participants and 23,954 incident cases. In both random-effects and fixed effects meta-analysis, butter consumption was associated with lower incidence of type 2 diabetes, with 4% lower risk per daily 14g serving: RR = 0.96 (95%CI = 0.93,0.99); P = 0.021). Moderate heterogeneity was seen (I² = 42.1%, p-heterogeneity = 0.131) (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Butter consumption and risk of type 2 diabetes.

Within-study dose-response RRs were derived from reported linear effects or generalized least-squares trend estimation for studies reporting categories of intake, a pooled using both inverse-variance weighted random and fixed effects meta-analysis. D+L: DerSimonian and Laird random effects; EPIC: European Prospective Investigation into Cancer and Nutrition; I-V: Inverse-variance fixed effects; RR (95%CI): Relative Risk and 95% Confidence Interval.

https://doi.org/10.1371/journal.pone.0158118.g004

Discussion

In this systematic review and meta-analysis of prospective studies, we found a small positive association between butter consumption and all-cause mortality, no significant association with incident CVD or CVD subtypes, and a modest inverse association with type 2 diabetesNo RCTs of butter intake were identified in our literature search. Because several of the identified reports included multiple country-specific cohorts, the total numbers of nation-specific cohorts, participants, and clinical events appear reasonably robust. Indeed, together these studies included more than 28,000 total deaths, nearly 10,000 cases of incident CVD, and nearly 24,000 cases of incident diabetes. We found limited formal evidence for between-study heterogeneity or publication bias, and all reports had high quality scores. Together, these findings suggest relatively small or neutral associations of butter consumption with long-term health.

Current dietary recommendations on butter and dairy fat are largely based upon predicted effects of specific individual nutrients (e.g., total saturated fat, calcium), rather than actual observed health effects. Our findings add to a growing body of evidence on long-term health effects of specific foods and types of fats. [12, 32, 33] Conventional guidelines on dietary fats have not accounted for their diverse food sources nor the specific individual fatty acid profiles in such foods. [4] Different foods represent complex matrices of nutrients, processing, and food structure, which together influence net health effects. [3, 34] Thus, studying intakes of foods, as in the present investigation, is crucial to elucidate health impact. Our novel results, together with other prior research described below, indicate a need for further funding, evaluation, and reporting on health effects of butter and dairy fat on mechanistic pathways and long-term health outcomes.

While prior meta-analyses have evaluated total dairy or some dairy subtypes and incident diabetes, to our knowledge none have evaluated butter and type 2 diabetes. [6, 12] A meta-analysis of butter and all-cause mortality identified no significant association (highest category vs. lowest: RR = 0.96; 95%CI = 0.95, 1.08) [10], but did not include the more recent large report from Sluik et al. [30](258,911 participants, 12,135 deaths) and also included two smaller studies not meeting our inclusion criteria: one having only crude (unadjusted) estimates, [35] and another evaluating polyunsaturated fats or margarine in comparison to butter, rather than butter separately. [36] A meta-analysis evaluating dairy consumption and CVD found no association between butter consumption and stroke (2 cohorts: RR = 0.94; 95%CI = 0.84, 1.06) or CHD (3 cohorts: RR = 1.02, 95%CI = 0.88, 1.20), but only evaluated high vs. low categories of intake rather than conducting dose-response analyses utilizing all available data. [5] Another meta-analysis included dose-response findings on butter consumption and stroke, but not CHD, CVD, diabetes, or all-cause mortality, and arrived at similar findings for stroke as seen in the present study. [11] In comparison to these prior reports, we evaluated up-to-date reports and full dose-response analyses for all-cause mortality, CVD including CHD and stroke, and type 2 diabetes; providing the most comprehensive investigation to-date of butter consumption and risk of long-term major health endpoints.

Our investigation also adds to and expands upon prior studies evaluating other dairy foods and dairy fat biomarkers in relation to cardiometabolic outcomes. In a multi-ethnic US population, serum levels of pentadecanoic acid (15:0), the odd-chain saturated fat most strongly associated with self-reported butter intakes (r = 0.13), were associated with lower CVD and CHD risk. [37] This is consistent with a meta-analysis of odd-chain saturated fat biomarkers demonstrating inverse associations with CHD [33]. A prior meta-analysis of dairy consumption and CVD suggested protective associations with total CVD (for highest vs lowest category of intake: 12% lower risk) and stroke (13% lower risk), with conflicting results for major subtypes of dairy. [5] Dairy fat has also been linked to lower risk of diabetes, based on studies of circulating fatty acid biomarkers [8, 9] and studies of self-reported consumption of dairy products, which have seen protective associations for yogurt and perhaps cheese, and null associations for both low-fat and whole-fat milk. [12, 27]

Given adverse effects of certain dairy fats (e.g. 16:0) on cardiometabolic risk factors such as LDL-cholesterol and fasting glucose [38, 39], our findings suggest potential presence of other mechanistic benefits of butter that might at least partly offset these harms. For instance, saturated fats also increase HDL-C, lower VLDL-C and chylomicron remnants, and lower lipoprotein(a) [40, 41]; while potential cardiometabolic benefits have been identified for calcium, fat-soluble vitamin D, medium-chain saturated fats, branched-chain fats, trace ruminant trans fats, or other processes related to fermentation (e.g. cheese) or active bacterial cultures (e.g. in yogurt). For example, dietary calcium may decrease fatty acid synthase and increase lipolytic activity in adipocytes, [42] reduce blood pressure by modulation of smooth muscle reactivity, [43, 44] and reduce weight gain. [45] Vitamin D may reduce dyslipidemia and improve blood pressure through maintenance of calcium homeostasis, stimulation of insulin production and release, and regulation of the renin-angiotensin-aldosterone system. Higher dairy fat consumption has been linked to lower liver fat and greater hepatic and systemic insulin sensitivity [46] which could relate to inhibition of hepatic de novo lipogenesis by specific dairy fatty acids. [8] Branched-chain fatty acids in dairy fat may promote healthier bacterial microbiome composition and function. Dairy fat also contains monounsaturated fats which might improve glycemic responses and insulin sensitivity. [47, 48] Other dairy-related factors, such as probiotic bacteria in yogurt and menaquinones in fermented milk and cheeses, may improve insulin sensitivity, reduce weight gain, and reduce inflammation through microbiome and vitamin-K related pathways; [49, 50] such pathways would be less relevant for butter, which has been linked to greater weight gain. [51, 52] Clearly, additional mechanistic studies on health effects of butter, dairy fat, and dairy foods are warranted.

Our results suggest relatively small or neutral overall associations of butter with mortality, CVD, and diabetes. These findings should be considered against clear harmful effects of refined grains, starches, and sugars on CVD and diabetes; [53–55] and corresponding benefits of fruits, nuts, legumes, n-6 rich vegetable oils, and possibly other foods such as fish on these endpoints. In sum, these results suggest that health effects of butter should be considered against the alternative choice. For instance, butter may be a more healthful choice than the white bread or potato on which it is commonly spread. In contrast, margarines, spreads, and cooking oils rich in healthful oils, such as soybean, canola, flaxseed, and extra-virgin olive oil, appear to be healthier choices than either butter or refined grains, starches, and sugars. [15, 56, 57] In Guasch-Ferre’s analysis of the Nurses Health Study, substitution of 8 g olive oil for an equivalent amount of butter was associated with an 8% reduction in the risk of type 2 diabetes (RR = 0.92 (95%CI = 0.87, 0.97). [15] Thus, even with an absence of major health associations in the present investigation, healthier (and less healthy) alternatives may be available. Our findings suggest a major focus on eating more or less butter, by itself, may not be linked to large differences in mortality, cardiovascular disease, or diabetes. In sum, our findings do not support a need for major emphasis in dietary guidelines on butter consumption, in comparison to other better established dietary priorities. In any meta-analysis, the effects of potential publication bias should be considered. Such bias increases the probability that large, positive associations, rather than small or null findings, will be published. In this case, the identified studies each reported generally modest or null findings. Considering the number of large prospective studies globally having data on dietary habits (including butter consumption) and these outcomes, it is evident that many additional cohort studies have collected such data but not analyzed or reported their findings. Such “missing,” unpublished studies may be more likely to have null effects. This may be particularly relevant for total mortality, with only 2 identified publications: additional publications might plausibly move findings toward the null. For diabetes, where a larger count of publications allowed better assessment for bias, the “trim and fill” method identified one theoretical missing study, with a protective point estimate.

Our investigation has several strengths. We followed stringent eligibility criteria that maximized inclusion of high quality, comparable studies. Our comprehensive literature search of multiple databases together with author contacts for clarification and missing data maximized statistical power and minimized the possibility of missed reports. While relatively few publications reported on certain outcomes, the identified studies were large, included multiple nation-specific cohorts and thousands of cases, and were of high quality; and as described above, it would be unlikely that publication bias would explain small or null (as opposed to large) associations. The inclusion of generally healthy participants followed since the 1980s and 1990s to the present provided populations generally free of lipid-lowering medications, which might otherwise mask full effects of butter on CVD. The identified cohorts provided a wide range of butter intakes, increasing power to detect an effect, if present. The dose-response analyses maximized use of all reported data, increasing precision.

Potential limitations should be considered. The health effect of any food could be modified by a person’s background diet, genetics, or risk factor profile. This is true for any lifestyle, pharmacologic, or other health intervention—effects may be modified by other treatments or underlying characteristics—but this does not lessen the relevance of evaluating the average population effect. We did not observe any obvious differences in associations based on country or region, where background dietary patterns might differ; but the number of identified studies precluded robust investigation of potential sources of heterogeneity. While the majority of studies adjusted for major demographic, clinical, and dietary covariates, residual confounding may be present. Because butter consumption is associated with generally worse diet patterns and lifestyle habits [58, 59], such residual confounding may overestimate potential harms of butter for mortality, and underestimate potential benefits of butter for CVD or diabetes. Error or bias in measurement of dietary intake from self-reports, as well as the long periods between dietary assessment and follow-up in several studies (10 years or more), may attenuate findings. On the other hand, even with such limitations, many other dietary factors in these and other cohorts have identified significant associations with mortality, CVD, and diabetes, so this is unlikely to be the sole explanation for the null findings. We did not identify any randomized clinical trials of our hard endpoints, although such a long-term trial focused on butter alone might be prohibitively expensive and impractical. Our results are based on best available observational findings, and long-term interventional studies were not found, limiting inference on causality.

In conclusion, the available evidence indicates small or neutral associations of butter consumption with all-cause mortality, CVD, and type 2 diabetes.

Supporting Information

Acknowledgments

We thank Dr Fumiaki Imamura (Medical Research Council Epidemiology, University of Cambridge), for his guidance in statistical analysis and Dr Renata Micha (Friedman School of Nutrition Science and Policy, Tufts University, Boston) for her comments on the study protocol. Drs Imamura and Micha did not receive any financial compensation for their assistance with this study. We thank Dr Sandra Goldbohm, Professor Frank Hu, Dr Susanna Larsson, Professor Paul Knekt, Dr Diewertje Sluik and Dr Emily Sonestedt for their correspondence and for providing clarifications and additional data on their respective studies.