Food Sources of Saturated Fat and the Association With Mortality: A Meta-Analysis

Abstract

We summarized the data related to foods high in saturated fat and risk of mortality. We searched Cochrane Library, MEDLINE, EMBASE, and ProQuest for studies from January 1952 to May 2012. We identified 26 publications with individual dietary data and all-cause, total cancer, or cardiovascular mortality as endpoints.

Pooled relative risk estimates demonstrated that high intakes of milk, cheese, yogurt, and butter were not associated with a significantly increased risk of mortality compared with low intakes. High intakes of meat and processed meat were significantly associated with an increased risk of mortality but were associated with a decreased risk in a subanalysis of Asian studies. The overall quality of studies was variable.

Associations varied by food group and population. This may be because of factors outside saturated fat content of individual foods. There is an ongoing need for improvement in assessment tools and methods that investigate food sources of saturated fat and mortality to inform dietary guidelines.

National dietary guidelines typically promote foods low in saturated fat.^1–3 These guidelines have arisen from early epidemiological studies showing that increased serum cholesterol was associated with increased risk of cardiovascular disease (CVD) and feeding studies showing that some, but not all, saturated fats increased serum cholesterol in comparison with unsaturated fats.^4,5 However, the effects of diet on CVD can be mediated through pathways other than total serum cholesterol or low-density lipoprotein cholesterol,⁶ and the use of intermediate measures such as cholesterol as outcomes could be misleading. Restriction of saturated fat is now being questioned,⁷ with a recent meta-analysis showing that intake of saturated fats was not significantly associated with an increased risk of CVD.⁸

As awareness of the relationship between diet, nutrition, and health increases in the general public, it is imperative that the dietary advice of health professionals be evidence based and reflect current scientific understanding. The recent debate regarding intake of saturated fats and risk of disease highlights 2 important questions for research.⁹ First, should dietary nutrients be considered in isolation? People consume foods, not individual nutrients. Thus, the effect of saturated fat needs to be considered in the context of its food sources. Individual saturated fatty acids may have different effects on mortality risk; for example, the type of saturated fat found in dairy products may be protective for chronic disease.^10–13 Second, are individual biological markers sufficient measures of risk compared with clinical endpoints such as mortality, which give a more definitive outcome? We conducted a meta-analysis of cohort studies reporting the relationship between key food groups typically high in saturated fatty acids and mortality in initially healthy adults. We tested the null hypothesis that there would be no significant association of saturated fat sources with all-cause, CVD, or cancer mortality.

METHODS

We searched the bibliographical databases of the Cochrane Library, MEDLINE, Embase, ProQuest, and ProQuest Dissertations and Theses for studies published in journals from January 1952 to May 2012. We adapted the search queries for use in each database and included alternative spellings of search terms where appropriate. We conducted searches using the following search algorithm (as formatted for Ovid): (dairy OR milk OR cheese OR yogurt OR butter OR meat OR animal flesh OR animal organ OR palm oil OR soybean oil OR cocoa OR chocolate OR lard OR tallow OR coconut) AND (mortality OR cancer OR coronary heart disease OR cardiovascular), limited to (human AND [adult < 18–64 years > OR aged < 65+ years > ]). We sought to identify additional studies, conference presentations, and theses by searching lists of relevant trials, reference lists from review articles, and meta-analysis and key articles. We did not contact researchers and field experts to obtain additional references. We considered studies published in any language for inclusion.

Two authors (T. O. S. and K. H.) assessed 506 studies for relevance and independently read titles, abstracts, and key words of identified articles. We obtained full-text versions for studies reporting mortality in relation to intakes of food types considered to be clinically relevant sources of naturally occurring saturated fat. Inclusion criteria included initially healthy human adult participants and dietary data reported in sufficient detail on an individual rather than a population basis. We chose mortality as the outcome, as it is generally well quantified and represents a final health outcome. Mortality type included all-cause, CVD, or cancer. Exclusion criteria included animal models and populations defined by preexisting disease or participants younger than 16 years. We did not place any restrictions on follow-up time. We included 26 studies after exclusions, representing data from 1 800 418 participants.^14–39 Figure 1 shows the flow of study selection for this meta-analysis.

FIGURE 1—

Flowchart of study inclusion: meta-analysis of the association of food sources of saturated fat with mortality: January 1952–May 2012.

Data Extraction

Two investigators (T. O. S. and K. H.) summarized studies, with any differences of opinion resolved after discussion and input from a third investigator as required. For studies meeting the inclusion criteria, we extracted relevant population and intervention characteristics with the use of standard extraction templates. Where results were reported for both high- and low-fat products for a food group (e.g., reduced fat milk and full cream milk or lean and fatty meats), we extracted the data for the high-fat product for inclusion in the analysis. For the meat category, we used results for red meat preferentially where specified. We extracted maximally adjusted results from each study. Principal summary measures were risk ratios (RRs) and hazard ratios (HRs).

Quality Assessment

We conducted the quality assessment to determine the likely risk of bias associated with each of the included studies and created a grading scheme to determine how appropriate the methods were in their ability to address our research hypothesis. Two authors (T. O. S. and K. H.) evaluated the included studies by using a modified quality assessment for cohort and case control studies derived from the Newcastle–Ottawa Quality Assessment Scale and checklists from Kmet et al.^40,41 We derived quality scores from sample size, participant dropout, and adequacy of description of missing participants; length of follow-up; dietary assessment method; outcome ascertainment method; degree of control for potential biases; description of analysis; and results reporting. Investigated studies differed in 3 main areas: length of follow-up, dietary assessment method, and degree of control for potential biases, such as smoking, physical activity, socioeconomic factors, body mass index (defined as weight in kilograms divided by the square of height in meters), and adjustment for energy intake. We graded studies as comprehensive, adequate, or limited for each of these main criteria:

• Sufficient length of follow-up in relation to age: youngest cohort members reach younger than  40 years = limited; 40–59 years = adequate;  60 years and older = comprehensive

• Dietary assessment quality: food frequency only, not quantitative or semiquantitative or meat eaters versus non–meat eaters = limited; quantitative or semiquantitative food frequency questionnaire or single 24-hour recall = adequate; diet record or diet history or multiple 24-hour recalls = comprehensive (provides more complete information and is better able to assess portion sizes and capture longer term variation⁴²)

• Multivariate analysis controlling for potential bias: basic age, gender = limited; at least 1 additional factor besides age and gender variables = adequate; at least 4 additional factors besides age and gender variables = comprehensive

We assigned an overall grade of comprehensive, adequate, somewhat limited, or limited overall on the basis of the criteria. We graded on the basis of the ability of each study to address our research aim. Our aim was to report the relationship between key food groups typically high in saturated fatty acids and risk of mortality, and the included studies could have mortality as a primary or secondary measure. Therefore, we derived quality assessment and scores from the level of detail regarding individual intakes of food groups of interest, the likelihood that the study follow-up was sufficient to include mortality outcomes, and the degree to which studies controlled for potential bias in the analysis.

Synthesis of Results

We used Review Manager 5.1⁴³ to complete the analyses unless otherwise specified. The main outcome measures were all-cause, CVD, and cancer mortality. We included results from the most complex model reported that controlled for the greatest number of confounding factors. We performed subgroup analyses for main food categories. We tested fixed and random effects models. Because of the large heterogeneity in results, as determined by the I² statistic, we have reported results for random effects models.

To examine potential dose–response and nonlinear relationships, we used the method Greenland and Longnecker described.⁴⁴ This method requires that the number of cases and total number of individuals be known for each category of intake. When these data were not provided, we approximated values from the number of person-years reported. Additionally, the mean or median level of intake is required for each category, and we estimated this value when it was not provided. For example, when density values such as grams of cheese per 1000 calories were reported, we converted values to grams per day using the average caloric intake of the relevant category. Or where intakes were reported as the number of servings consumed in a given time period, we converted these values to number of grams per day using the following serving sizes: meat 120 grams, processed meat 50 grams, milk 200 milliliters, yogurt 200 grams, cheese 40 grams, total dairy 200 grams, and butter 10 grams. For each food type, we determined the risk associated with each additional serving consumed per week. We used random effects models. To test for nonlinearity, we assessed the difference between the nonlinear and linear models. We completed all dose–response analyses in SAS 9.2.⁴⁵

RESULTS

Key characteristics of the 26 included studies are displayed in Table 1. All were prospective observational studies. Populations were geographically diverse, with the majority of studies from the United States (n = 12) and Japan (n = 7). There were substantial differences in cohort size, with samples varying between 162 and 764 343. Study durations varied between 5 and 41 years, and participant age at enrollment ranged from 15 through 103 years. Most studies included both male and female cohorts.

TABLE 1—

Baseline Characteristics of Participants in 26 Prospective Cohort Studies of High–Saturated Fat Food Types and Mortality, Meta-Analysis of the Association of Food Sources of Saturated Fat With Mortality: January 1952–May 2012

Study	Participants, No.	Age, Years	Follow-up, Years	Gender	Mortality Cause (No. Cases)	Food Type	Participant Group, Country of Residence
Bonthuis et al.14	1529	25–78	16	Both	All-cause (n = 177), CVD (n = 61)	Dairy, milk, cheese, yogurt	Nambour Skin Cancer Prevention Trial, Australia
Bostick et al.15	34 486	55–69	8	Female	CVD (n = 387)	Dairy, milk	Iowa Women’s Health Study, United States
Fortes et al.16	162	≥ 65	5	Both	All-cause (n = 53)	Meat, dairy, milk, cheese, butter	Home for the Elderly, Rome, Italy
Fraser and Shavlik17	603	≥ 85	12	Both	All-cause (n = 1387), CVD (n = 364)	Cheese	The Adventist Health Study
Goldbohm et al.18	20 782	55–69	10	Both	All-cause (n = 16 136), CVD (n = 4288)	Milk, cheese, butter	Netherlands Cohort Study, Netherlands
Kinjo et al.19	223 170	40–69	15	Both	CVD (n = 11 030)	Meat, milk	Japan
Kojima et al.20	107 824	40–79	10	Both	Cancer (n = 284 colon, n = 173 rectal)	Meat, processed meat, milk, yogurt, cheese, butter	Japan Collaborative Cohort Study, Japan
Mann et al.21	10 802	16–79	13	Both	All-cause (n = 392), IHD (n = 64)	Meat, eggs, milk, cheese	Vegetarian Society and Family/Friends, United Kingdom
Matsumoto et al.22	11 606	18–90	9	Both	Cancer (n = 255)	Milk, yogurt, butter	Jichi Medical School Cohort Study, Japan
Mills et al.23	994	30–85	8	Female	Cancer (n = 142; breast)	Meat, eggs, milk, cheese	Seventh-Day Adventists, United States
Mills et al.24	34 198	25–≥ 95	6	Both	Cancer (n = 40; pancreatic)	Meat, eggs, milk, cheese	Seventh-Day Adventists, United States
Ngoan et al.25	13 250	15–96	13	Both	Cancer (n = 116; stomach)	Meat, processed meat, milk	Fukuoka Prefecture, Japan
Ozasa et al.26	98 248	40–79	9	Both	Cancer (n = 527; lung)	Processed meat, milk, yogurt, cheese, butter	Japan Collaborative Cohort Study, Japan
Pan et al.27	121 342	30–75	22–28	Both	All-cause (n = 23 926), cancer (n = 9464), CVD (n = 5910)	Meat, processed meat	Health Professionals Follow-Up Study and Nurses’ Health Study, United States
Park et al.28	293 888	50–71	6	Male	Cancer (n = 178; prostate)	Dairy, milk, yogurt, cheese	National Institutes of Health–AARP, United States
Phillips and Snowdon29	25 493	≥ 35	21	Both	Cancer (n = 182; colorectal)	Meat, dairy, milk	Seventh-Day Adventists, United States
Qiu et al.30	50 069	40–≥ 80	6	Both	CVD (n = 632; CHD)	Meat	Rural China
Sakauchi et al.31	114 517	40–≥ 80	10	Both	Cancer (n = 85; urothelial)	Meat, processed meat, milk, yogurt, cheese, butter	Japan Collaborative Cohort Study, Japan
Sauvaget et al.32	37 130	34–103	16	Both	CVD (n = 1462; stroke)	Meat, processed meat, dairy, milk	Hiroshima/Nagasaki Lifespan Study, Japan
Sinha et al.33	545 653	50–71	10	Both	All-cause (n = 71 252), cancer (n = 25 362), CVD (n = 19 577)	Meat, processed meat	National Institutes of Health–AARP, United States
Smit et al.34	977	35–79	41	Male	Cancer (n = 167; prostate)	Meat, dairy	Puerto Rico Heart Health Program, United States
Snowdon et al.35	6763	≥ 39	21	Male	Cancer (n = 99; prostate)	Meat, milk, cheese	Seventh-Day Adventists, United States
Thorogood et al.36	11 130	39a	12	Both	All-cause (n = 200), cancer (n = 164), CVD (n = 94; IHD)	Meat	Vegetarian Society and Family/Friends, United Kingdom
Thun et al.37	764 343	≥ 30	6	Both	Cancer (n = 1150; colon)	Meat	Cancer Prevention Study II, United States
Whiteman et al.38	10 522	35–64	9	Both	All-cause (n = 514), cancer (n = 235), CVD (n = 107)	Meat, processed meat, milk, butter	OXCHECK, United Kingdom
Zheng et al.39	17 633	≥ 35	20	Male	Cancer (n = 57; pancreatic)	Meat	Lutheran Brotherhood Insurance Policyholders, United States

Note. CHD = coronary heart disease; CVD = cardiovascular disease; IHD = ischemic heart disease.

^aMean age.

Total mortality was reported in 9 studies. Additionally, 5 reported total cancer, and 12 reported CVD-related mortality outcomes. Nine studies reported specific cancer mortality data. Some reported multiple outcomes of interest (Table 1).

Non- or semiquantitative food frequency questionnaires were used to assess diet in 24 of the 26 studies. The number of items assessed by the questionnaires ranged from 8 to 166, with 9 studies assessing portion size in the questionnaires. One study compared vegetarian with nonvegetarian, and another used a single 24-hour recall. None of the included studies used multiple-day diet records or recalls. A description of the quality of included studies for the purpose of this analysis is displayed in Table 2. We were able to investigate relative risk of mortality in the meta-analysis for food groups of meat, processed meat, milk, butter, cheese, and dairy as a whole.

TABLE 2—

Quality Assessment of Studies Included: Meta-Analysis of the Association of Food Sources of Saturated Fat With Mortality: January 1952–May 2012

	Follow-Up/Youngest Participant Age, Yearsa		Dietary Assessmentb		Confounding Factors Consideredc (in Addition to Age and Gender)
Study	Study Data	Study Grade	Study Data	Study Grade	Study Data	Study Grade	Overall Grade
Comprehensive
Bostick et al.15	8/55	LL	127-item SQ FFQ, validated weighed food records (dairy r = 0.75)	L	BMI, waist:hip, medical history, smoking, alcohol, education, physical activity, other dietary factors	LL	LL
Fortes et al.16	5/65	LL	114-item SQ FFQ, validated vs 7-d weighed record	L	BMI, smoking, cognitive status, medical history, education	LL	LL
Goldbohm et al.18	10/55	LL	150-item SQ FFQ, validated 9-d diet record (milk r = 0.60; cheese r = 0.61)	L	Education, smoking, BMI, physical activity, vitamin use, alcohol, other dietary factors	LL	LL
Sinha et al.33	10/50	LL	124-item SQ FFQ, validated 2 × 24-hr recalls (iron r = 0.59 male and 0.56 female, not validated on food group level)	L	Smoking, physical activity, education, marital status, medical history, race, BMI, alcohol, supplements, HRT, other dietary factors	LL	LL
Smit et al.34	41/35	LL	1 × 24-hr recall, quantitative	L	Smoking, BMI, physical activity, urban living, other dietary factors	LL	LL
Adequate
Bonthuis et al.14	16/25	L	129-item SQ FFQ, validated 5 × 24-hr recall (calcium r = 0.67; reliability ranged r = 0.57–0.82)	L	BMI, alcohol, education, physical activity, smoking, supplements, medical history, occupation, other dietary factors	LL	L
Fraser et al.17	12/84	LL	65-item FFQ	XXX	BMI, education, physical activity, medical history, smoking	LL	L
Kojima et al.20	10/40	L	33-item FFQ validated 4 × 3-d food record (all studied foods > r = 0.3; reproducibility ranged r = 0.4–0.8)	XXX	Physical activity, education, medical history, BMI, alcohol, smoking	LL	L
Pan et al.27	22–28/30	L	131–166-item SQ FFQ, validated 4 × 7-d diet record (r = 0.59 unprocessed red meat, 0.56 processed red meat); mean 12-month reliability (r = 0.57)	L	BMI, alcohol, physical activity, smoking, race, menopausal status and hormone use, family history, medical history, other dietary factors	LL	L
Park et al.28	6/50	L	124-item SQ FFQ, validated 2 × 24-hr recall (calcium r = 0.63, not validated on food group level)	L	Race, education, marital status, BMI, physical activity, smoking, alcohol, medical history, other dietary factors	LL	L
Qiu et al.30	6/40	L	10-item FFQ	XXX	Regional area, smoking, alcohol, medical history, BMI, marital status, sleep, other dietary factors	LL	L
Sauvaget et al.32	16/34	L	22-item FFQ, validated 1 × 24-hr diet record (range r = 0.17–0.32)	XXX	Smoking, BMI, education, medical history, radiation, regional area	LL	L
Snowdon et al.35	21/39	LL	FFQ, 1 food group per Q	XXX	Education, % desirable wt, other dietary factors	L	L
Somewhat limited
Kinjo et al.19	15/40	L	FFQ, 1 food group per Q (meat, milk, fish)	XXX	Occupation, smoking, alcohol	L	1/2
Mann et al.21	13/16	XXX	SQ FFQ, 1 food group per Q, validity done for fiber only	L	Smoking, social class	L	1/2
Mills et al.23	8/30	XXX	21-item FFQ	XXX	Age at menarche, first pregnancy, and menopause, % desirable wt, education	LL	1/2
Ngoan et al.25	13/15	XXX	25-item FFQ	XXX	Smoking, alcohol, medical, occupation, coffee	LL	1/2
Ozasa et al.26	9/40	L	32-item FFQ, validated 4 × 3-d diet records (milk, ham, and sausage: r > 0.6)	XXX	Family history, smoking	L	1/2
Phillips et al.29	21/35	L	21-item FFQ	XXX	Coffee, % desirable wt, other dietary factors	L	1/2
Sakauchi et al.31	10/40	L	32-item FFQ, validated 4 × 3-d diet records (milk, ham, and sausage: r > 0.6)	XXX	Smoking	L	1/2
Thun et al.37	6/30	XXX	32-item FFQ	XXX	Family history, physical activity, BMI, aspirin use, other dietary factors	LL	1/2
Whiteman et al.38	9/35	L	8-item FFQ	XXX	Smoking, alcohol, SES	L	1/2
Zheng et al.39	20/35	L	35-item FFQ	XXX	Smoking, alcohol, energy intake	L	1/2
Limited
Mills et al.24	6/25	XXX	FFQ past or current use by food group	XXX	Smoking	L	XXX
Thorogood et al.36	12/39e	XXX	Nonmeat eaters vs meat eaters	XXX	Smoking, BMI, SES	L	XXX
Matsumoto et al.22	9/18	XXX	30-item FFQ, validated 12-d diet records by other study that used same FFQ but “included more food items” (r = 0.65 milk)	XXX	…	XXX	XXX

Note. BMI = body mass index; FFQ = food frequency questionnaire; HRT = hormone replacement therapy; SES = socioeconomic status; SQ = semiquantitative; Q = question, wt = weight. We derived grading from the ability of each study to address our research aims; it did not necessarily reflect the quality of research studies we investigated. We graded for each category of interest according to rankings of XXX = limited; L = adequate; LL = comprehensive. Total ranking represents average of the category grading and includes 1/2 = somewhat limited.

^bSufficient length of follow-up in relation to age: youngest cohort members reach < 40 years = XXX; 40–59 years = L; ≥ 60 years = LL.

^cDietary assessment quality: food frequency only, not quantitative or semiquantitative = XXX; quantitative or semiquantitative food frequency questionnaire or meat eaters versus nonmeat eaters or single 24-h recall = L; diet record or diet history or multiple 24-h recalls = LL.

^dMultivariate analysis controlling for potential bias: basic age, gender = XXX; at least 1 additional factor besides age and gender variables = L; at least 4 additional factors besides age and gender variables = LL.

^eMean age.

Although all studies adjusted for age and gender when applicable, there was substantial variation in the degree of adjustment for potential confounding factors, such as level of physical activity, smoking, body mass index, socioeconomic factors, and alcohol consumption (Table 2).

Meta-Analysis

Individual study results and pooled estimates for all-cause mortality are provided in Figure 2. There was no significant relationship between intake of milk, cheese, butter, or all dairy and all-cause mortality. High intake of meat (RR = 1.17; 95% confidence interval [CI] = 1.08, 1.27) and processed meat (RR = 1.21; 95% CI = 1.16, 1.28) were significantly associated with an increased risk of all-cause mortality. Substantial heterogeneity was evident in studies reporting risk of mortality and consumption of meat (I² = 85%), processed meat (I² = 65%), and butter (I² = 78%).

FIGURE 2—

High consumption of foods containing saturated fat and the associated risk of all-cause mortality: meta-analysis of the association of food sources of saturated fat with mortality: January 1952–May 2012.

Note. CI = confidence interval; F = female; M = male.

As displayed in Figure 3, there were no significant associations of high intake of meat, milk, cheese, or all dairy products with CVD mortality. However, high intake of processed meat was significantly associated with increased risk of CVD mortality (RR = 1.17; 95% CI = 1.02, 1.33). High heterogeneity existed between studies relating to consumption of meat (I² = 93%), processed meat (I² = 88%), and milk (I² = 82%).

FIGURE 3—

High consumption of foods containing saturated fat and the associated risk of cardiovascular disease mortality: meta-analysis of the association of food sources of saturated fat with mortality: January 1952–May 2012.

Meat, processed meat, and milk were the only food groups with data from more than 1 study for all-cancer mortality. For individual food groups, high consumption of meat (RR = 1.14; 95% CI = 1.04, 1.24) and processed meat (RR = 1.13; 95% CI = 1.09, 1.17) were associated with an increased risk of cancer death; milk was not (Figure 4).

FIGURE 4—

High consumption of foods containing saturated fat and the associated risk of cancer mortality: meta-analysis of the association of food sources of saturated fat with mortality: January 1952–May 2012.

A significantly increased risk of specific cancer mortality was associated with high compared with low meat intake (RR = 1.25; 95% CI = 1.03, 1.52) but not processed meat (RR = 1.25; 95% CI = 0.91, 1.72) or milk (RR = 0.92; 95% CI = 0.79, 1.09). Moderate heterogeneity existed for studies relating to meat consumption (I² = 60%; Figure A, available as a supplement to this article at http://www.ajph.org).

Subgroup Analysis

Subgroup analysis was limited because of the small number of studies in each food group and outcome. However, we examined the effect of eliminating studies classified as being of limited and somewhat limited quality. In addition, because of the large differences in the classification of meat intake in Asian and non-Asian populations, we conducted analyses separately for these 2 groups.

In terms of study quality, we graded 5 of 26 studies as being able to comprehensively address our research aim, whereas, at the other end of the spectrum, we graded 3 as limited (Table 2). We investigated changes in results when we excluded the limited studies from analysis. These studies were Matsumoto et al.,⁴⁶ who investigated cancer mortality; Thorogood et al.,³⁶ who examined all-cause mortality; and Mills et al.,²⁴ who investigated pancreatic cancer mortality. Exclusion of the Matsumoto et al.⁴⁶ study from the pooled analysis did not significantly change the results (RR = 1.14; 95% CI = 1.09, 1.19; I² = 57%), nor did exclusion of Thorogood et al.³⁶ (RR = 1.16; 95% CI = 1.13, 1.19; I² = 3%), or the exclusion of both (RR = 1.16; 95% CI = 1.13, 1.19; I² = 9%). For risk of specific cancer mortality, exclusion of the Mills et al. study²⁴ resulted in a nonsignificant relationship between meat intake and mortality (RR = 1.20; 95% CI = 0.96, 1.50; I² = 9%).

To determine whether differences in study quality were contributing to the significant heterogeneity evident in results, we removed those studies we considered of limited quality. In an additional step, we removed the studies we identified as of limited and somewhat limited quality to determine the impact on results. Although heterogeneity was not substantially reduced for most analyses, it was reduced for analysis of the relationship between meat and all-cause (I² = 30.4%), all-cancer (I² = 0%), and CVD (I² = 73%) mortality as well as milk and CVD mortality (I² = 0%).

The pooled analysis of non-Asian cohorts^21,33,38 suggested a significantly increased risk of CVD mortality with high intake of meat (RR = 1.29; 95% CI = 1.17, 1.42; I² = 70%). Conversely, studies of Asian populations^19,30,32 demonstrated a significantly reduced risk of CVD mortality (RR = 0.82; 95% CI = 0.73, 0.91; I² = 19%). Elimination of the 1 Asian cohort for processed meat did not change the direction of results with CVD mortality (RR = 1.24; 95% CI = 1.10, 1.40; I² = 82%). There were no significant associations of the high intake of milk with CVD mortality in either Asian or non-Asian populations (RR = 0.84; 95% CI = 0.72, 1.00; I² = 72% and RR = 1.07; 95% CI = 0.98, 1.17; I² = 0%, respectively).

Nonlinear and Dose–Response Relationships

Where sufficient data were available, we investigated potential nonlinear and dose–response relationships between the number of weekly servings of meat, processed meat, milk, and cheese and risk of all-cause, cardiovascular, and cancer mortality (Table 3). We observed a significant positive linear relationship between risk of all-cause mortality and consumption of meat and processed meat (Figure B, available as a supplement to this article at http://www.ajph.org). We also observed a significant negative curvilinear relationship between number of servings of milk consumed per week and risk of all-cause mortality, but the decrease in risk was minimal (Figure C, available as a supplement to this article at http://www.ajph.org).

TABLE 3—

Mortality Risk Associated With Each Additional Serving of Food per Week, by Food Category and Mortality Outcome: Meta-Analysis of the Association of Food Sources of Saturated Fat With Mortality: January 1952–May 2012

Food Type (Serving Size)	RR (95% CI) Linear Relationship	Studies, No.	Nonlinear Relationshipa	Heterogeneity (τ2)
All-cause mortality
Milk (200 mL)	1.00 (1.00, 1.00)	4	0.036*	< 0.001
Meat (120 g)	1.02* (1.01, 1.04)	7	0.007*	< 0.001
Processed meat (50 g)	1.04* (1.00, 1.07)	2	0.139	0.001
Cheeseb (40 g)	1.00 (1.00, 1.00)	5	0.106	< 0.001
CVD mortality
Milk (200 mL)	1.00 (0.99, 1.01)	5	0.014*	0.001
Meat (120 g)	1.01 (0.99, 1.04)	9	0.323	0.001
Processed meatb (50 g)	1.02* (1.01, 1.03)	2	0.005*	< 0.001
Cheese (40 g)	1.01 (0.99, 1.02)	3	0.013*	0.001
Cancer mortality
Meat (120 g)	1.02* (1.02, 1.03)	4	0.211	< 0.001
Processed meat (50 g)	1.04 (1.00, 1.08)	2	< 0.001*	0.001

Note. CI = confidence interval; CVD = cardiovascular disease; RR = risk ratio.

^aTest of significant departure from linear.

^bFixed effects and random effects results are the same trend.

^*P < .05.

There was no significant relationship between number of servings of cheese consumed per week and mortality. Significant curvilinear relationships existed between CVD mortality and intake of milk (negative association; Figure D, available as a supplement to this article at http://www.ajph.org), cheese (Figure E, available as a supplement to this article at http://www.ajph.org), and processed meat (Figure F, available as a supplement to this article at http://www.ajph.org). We observed the greatest reduction in risk for milk around 12 servings per week. Cheese displayed an incline in risk from around 8 servings per week, and processed meat from around 3 servings per week. There was a significant positive relationship between intake of meat and cancer mortality (Figure G, available as a supplement to this article at http://www.ajph.org).

Because of the small number of studies in each analysis, there was limited statistical power to detect publication bias; therefore, we assessed bias with the use of funnel plots. We observed little evidence of bias (Figure 5).

FIGURE 5—

Funnel plot of publication bias for all-cause mortality, by food group: meta-analysis of the association of food sources of saturated fat with mortality: January 1952–May 2012.

DISCUSSION

Early guidelines regarding intake of saturated fat, which remain largely unchanged today, lacked a strong evidence base. These guidelines were derived from assumptions about the causal pathway of disease development, specifically that replacing saturated fat in the diet with polyunsaturated fat lowered total serum cholesterol, which in turn reduced risk of CVD.^47,48 This is now thought to be an oversimplification of the multiple processes that influence the impact of dietary composition on human health. Our review, which summarizes the currently available evidence, is unable to support a strong recommendation regarding restricting intake of foods high in saturated fat for the prevention of mortality.

The results of our review suggest milk, cheese, butter, and total dairy intake were not significantly associated with increased risk of mortality. By contrast, higher intakes of meat and processed meat were significantly associated with modest increased risk of all-cause and cancer mortality, and higher intakes of processed meat were also significantly associated with a modest increase in CVD mortality. We observed dose–response relationships for meat and processed meat across mortality categories, with the majority displaying a curvilinear relationship with increasing risk at higher intakes. The total relative risk estimate for all-cause mortality for all included food groups suggested a small increase in risk with higher intakes (RR = 1.09; 95% CI = 1.03, 1.14). However, considering the low quality of most of the included studies and the large variation in the results between individual food groups, we were unable to provide support for, or refute, the existing guidelines regarding consumption of saturated fat. These results highlight the need for further high-quality research regarding intake of foods high in saturated fats and disease outcomes before dietary guidelines or public health recommendations are proposed or implemented.

The positive association observed between meat intake and mortality risk agrees with those previously reported in relation to cancer risk. A meta-analysis of almost 8000 cases from 19 prospective studies found consumption of red meat and processed meat to be linked with risk of developing both colon and rectal cancer.⁴⁹ However, research investigating mechanisms behind this result suggests that these associations are likely to be because of factors outside saturated fat content. Unlike dairy products, meat contains high amounts of heme iron, which has been shown to damage the colonic mucosa in rat models, resulting in a hyperproliferation of the epithelium, which may have carcinogenic effects.⁵⁰ A prospective cohort study supports this concept, showing intake of heme iron from red meat was positively associated with risk of CVD, whereas non–heme iron intake was not significant.⁵¹ Nitrosation may also increase the toxicity of heme in cured products such as processed meats.⁵² Cooking meat at high temperatures is also thought to result in other potential mutagens and carcinogens such as heterocyclic amines and polycyclic aromatic hydrocarbons.⁵² Notably, our results showed a significantly reduced risk of CVD mortality associated with high compared with low consumption of meat in Asian populations, who are reported to have lower intakes of saturated fat than do Western populations.⁵³ Meat intake may modify the effects of saturated fat on atherogenic lipoproteins, as suggested by results of a diet trial that showed that replacing carbohydrate with beef resulted in improvements when combined with diets that were overall low in saturated fat but not high in saturated fat.⁵⁴ Therefore, the dietary context of foods may also be important along with the foods themselves. Another potential explanation is that the population subgroup differences may be a result of disease etiology rather than meat intake. Studies that reported outcomes for stroke and cerebrovascular disease predominately examined Asian populations, whereas those that reported cardiovascular outcomes examined non-Asian cohorts.

The meta-analysis by Siri-Tarino et al.⁸ found no significant associations with total saturated fat intake and risk of CVD. This research relates directly to recommendations to limit saturated fat as a whole in the diet. De Oliveira Otto et al.⁵⁵ used a food-based approach and reported that a higher intake of dairy saturated fat was associated with lower CVD risk, whereas a higher intake of meat saturated fat was associated with higher risk. The authors state that associations between saturated fat and health may depend on food-specific fatty acids or other nutrient constituents in addition to saturated fat. Taken together with our findings, it appears that the role of saturated fat in health may differ on the basis of the source and type of saturated fat consumed rather than on the total amount.

It has been suggested that it is not feasible to separate different types of saturated fat with respect to food choices because the foods contain a combination of several different saturated fatty acids.⁹ The types of saturated fat in meat, processed meat, milk, yogurt, cheese, and butter are predominantly palmitic acid (16:0), followed by stearic acid (18:0), and then myristic acid (14:0; Table 4). Both stearic acid and palmitic acid, the predominant saturated fatty acids in the food types studied, have been shown to have a neutral to favorable impact on serum lipid profiles compared with lauric acid and myristic acid.⁵⁷ Stearic acid seems to be more beneficial than is palmitic acid when the 2 are compared,⁵⁸ and stearic acid but not trans fatty acid has been shown to significantly reduce concentrations of low-density lipoprotein cholesterol.⁵⁹ Although there is not yet enough evidence to give dietary recommendations for individual saturated fatty acids, our research raises questions regarding the traditional belief that all high–saturated fat foods lead to increased mortality. As people consume foods as a whole rather than individual nutrients, food-based dietary advice (i.e., consuming more or less of particular foods) rather than nutrient-based dietary advice (i.e., consuming more or less of individual nutrients) may be more practical for health professionals to communicate to the public.

TABLE 4—

Saturated Fatty Acid Profile of Food Types High in Naturally Occurring Saturated Fat: Meta-Analysis of the Association of Food Sources of Saturated Fat With Mortality: January 1952–May 2012

Food Type	Saturated Fat Content, g/100 g	Predominant Saturated Fatty Acid, Type (g/100 g)	Other Major Contributing Saturated Fatty Acids, Type (g/100 g)
Meata	7.2	16:0 (4.1)	18:0 (2.0), 14:0 (0.6)
Processed meatb	11.6	16:0 (7.1)	18:0 (3.9), 14:0 (0.5)
Milkc	1.9	16:0 (0.8)	18:0 (0.4), 14:0 (0.3)
Yogurtd	2.1	16:0 (0.9)	14:0 (0.3), 18:0 (0.3)
Cheesee	21.1	16:0 (9.8)	18:0 (4.0), 14:0 (3.3)
Butterf	51.4	16:0 (21.7)	18:0 (10.0), 14:0 (7.4)

Source. US Department of Agrictulture.⁵⁶

^aDerived from beef, ground, 70% lean meat or 30% fat, cooked.

^bDerived from luncheon meat, pork, beef.

^cDerived from milk, whole, 3.25% milk fat.

^dDerived from yogurt, plain, whole milk.

^eDerived from cheese, cheddar.

^fDerived from butter, without salt.

Limitations

Strengths of our study include the use of the relatively well-measured and decisive measure of mortality as an outcome, the inclusion of many large-scale and diverse studies, and the evaluation of food groups rather than nutrients. However, the small number of studies in some food categories was a limitation for evaluating associations of mortality with certain foods. A further limitation was the wide variation in the quality of dietary assessment tools. Food frequency questionnaires, which almost all studies included in our analysis used, are useful for assessing large cohorts, but some did not attempt to directly quantify serving size, were not validated with food group–specific intake, or reported correlation coefficients as low as 0.17. The degree of accounting for potential confounding factors studies used also varied.

Appropriate adjustment for confounding factors is important considering that a healthy lifestyle effect may have existed in some studies, whereby saturated fat would generally be limited with healthier lifestyles in accordance with dietary guidelines or those with existing disease or a family history of heart disease. This is also true for studies of populations who have a relatively strict dietary and lifestyle approach, such as Seventh-Day Adventists. Although our focus was on foods providing saturated fat in the diet, it was not possible to extract the effect of saturated fat from other nutrients in a food, and some nutrients present in food we investigated may obscure the relationship between saturated fat and mortality. On the other hand, some studies used models including numerous factors, thereby increasing the possibility of overadjustment, resulting in loss of power to see an association where a true association exists. In the absence of any randomized experiments, observational studies provide the best evidence we have available. However, these weaknesses are inherent to observational epidemiology as a field and should be considered in the interpretation of our results.

A potential criticism of our study is that deriving a quantitative estimate of risk though meta-analysis is not appropriate because of the quality deficiencies and substantial differences between the studies included in this analysis (Table 1). Although results should be interpreted in the context of this heterogeneity, we feel that there is value in combining study results to provide the best quantitative assessment of the current evidence because of the existence of longstanding and specific dietary guidelines related to saturated fat consumption. On the basis of our meta-analysis we are unable to provide support for or propose changes to current guidelines and recommendations. Instead, this review highlights the need for high-quality research regarding intakes of saturated fats, in particular regarding food sources of saturated fats and disease outcomes.

Most dietary guidelines recommend limiting saturated fat to 10% or less of total energy intake; however, these guidelines generally do not specify the replacement nutrient.⁹ Replacing saturated fats with polyunsaturated fats is considered to be beneficial to health⁷: Jakobsen et al.⁶⁰ demonstrated a significant inverse association of polyunsaturated fats with risk of coronary events in a pooled analysis of cohort studies; Mozafarrian et al.⁶¹ showed that consumption of polyunsaturated fat in place of saturated fat reduces coronary heart disease events in randomized controlled trials. Furthermore, Jakobsen et al. suggested that replacing high–saturated fat foods in the diet with refined carbohydrate foods may be detrimental to health.⁶⁰ Therefore, it may be more in keeping with available evidence if guidelines say to either replace saturated fats with polyunsaturated fats or focus more on food-based recommendations.

Conclusions

Our study reveals the ongoing need for improvement in assessment tools and methods that investigate food sources of saturated fat and mortality. We believe the dietary assessment tools for such studies would be improved through the use of more comprehensive estimation of diet in the form of multiple 24-hour recalls, diet histories, or diet records rather than the heavy reliance on food frequency questionnaires, which often do not assess portion size and may not be well validated on a specific food group level. Despite the central role that dietary advice plays in public health promotion and prevention, our quality assessment identified a lack of comprehensive studies in the area of mortality and foods high in saturated fat. Studies that use quantitative dietary assessments, with a sufficient length of follow-up and sufficient personal and lifestyle data to adequately adjust for potential confounders, would be a welcome addition to the literature and bring needed clarity to the scientific evidence of the association of saturated fat with health.