The aim of this systematic review and meta-analysis was to summarize the evidence from observational studies assessing the association between intake of trans fatty acids (TFA) and the risk of coronary heart disease (CHD), with a specific emphasis on distinguishing between TFA of industrial and ruminant origin. By searching five bibliographic databases, analyses from six published and two unpublished prospective cohort studies, assessing the association of intake of TFA with fatal and/or non-fatal CHD, were identified. Four and three studies reported separate associations for intake of ruminant or industrial-TFA, respectively. The pooled relative risk estimates for comparison of extreme quintiles of total-TFA intake (corresponding to intake increments ranging from 2.8 to ∼10 g/day) were 1.22 (95% confidence interval: 1.08–1.38; P=0.002) for CHD events and 1.24 (1.07–1.43; P=0.003) for fatal CHD. Ruminant-TFA intake (increments ranging from 0.5 to 1.9 g/day) was not significantly associated with risk of CHD (risk ratio (RR)=0.92 (0.76–1.11); P=0.36), and neither was industrial-TFA intake, although there was a trend towards a positive association (RR=1.21 (0.97–1.50); P=0.09). In conclusion, our analysis suggests that industrial-TFA may be positively related to CHD, whereas ruminant-TFA is not, but the limited number of available studies prohibits any firm conclusions concerning whether the source of TFA is important. The null association of ruminant-TFA with CHD risk may be due to lower intake levels.
Dietary trans fatty acids (TFA) have long been suspected of increasing the risk of coronary heart disease (CHD), especially at high intake levels (Willett et al., 1993). No randomized intervention studies evaluating the effect of dietary TFA on hard end points for CHD have been conducted. However, as several epidemiological studies have shown a strong positive association between the intake of TFA and risk of CHD (Pietinen et al., 1997; Oomen et al., 2001; Oh et al., 2005), the suspicion of a harmful effect of TFA seems justified. It has been estimated that a 2% increase in energy intake from TFA is associated with a 23% increased risk of CHD in a previous pooled analysis of four prospective cohort studies with nearly 140 000 subjects (Mozaffarian et al., 2006).
However, TFA is not just TFA. TFA in food originate from two main sources; industrial partial hydrogenation of edible oils, and bacterial hydrogenation of unsaturated fatty acids in the rumen of ruminants. Today, the industrially produced TFA (IP-TFA) are primarily found in snacks and fast food products whereas the ruminant-TFA (R-TFA) constitute a part of the fat in meat and dairy products. The isomer distribution of the TFA from these two sources is dissimilar. In R-TFA the predominant isomer is vaccenic acid (t18:1n-7), whereas IP-TFA generally has a Gaussian distribution of t18:1 isomers with highest levels of elaidic acid (t18:1n-9) (Wolff et al., 1998).
There has been debate as to whether R-TFA is equally harmful as IP-TFA. Some results from epidemiological studies of intake of R-TFA and risk of CHD have indicated that intake of R-TFA is innocuous or even protects against CHD (Jakobsen et al., 2008).
We found it fitting and timely to conduct a systematic review and meta-analysis to assess the empirical evidence of an association between intake of TFA and the risk of fatal and/or non-fatal CHD, with a specific emphasis on stratifying results according to industrial or ruminant origin of the TFA. We aimed for inclusion of all the available prospective cohort studies, which have assessed the association between intake of TFA and incident CHD in adult populations.
Subjects and methods
We conducted this review in accordance with the PRISMA Statement for Reporting Systematic Reviews and Meta-Analyses of Studies That Evaluate Health Care Interventions (Liberati et al., 2009) on the basis of a predefined protocol, which was made publicly available and distributed among interested parties.
With this work, we aimed to review all prospective cohort studies describing the association between intake of TFA (industrial and/or ruminant) and the incidence of fatal and non-fatal CHD. We conducted a systematic literature search of MEDLINE via Pubmed from 1950 to March 2010; EMBASE via Ovid from 1980 to March 2010; Food Science and Technology Abstracts via Ovid from 1969 to March 2010; Web of Science from 1900-14 to March 2010, and SciFinder Scholar for studies describing the association between TFA intake and incidence of CHD. Two search themes were combined: TFA intake and CHD mortality and morbidity (for detailed description of the search strategy please refer to Supplemental Appendix A). The search strategy had no language or study design restrictions. Instead, prospective cohort studies were identified from titles and abstracts of the retrieved references. Additionally, we screened the reference lists of all identified relevant studies, and of review articles, in order to identify possible studies of interest. We contacted authors of included studies for references to studies not identifiable by our other searches, including unpublished studies.
Two reviewers (NTB, investigator, and EMB, Senior Researcher in Biophysics and Research Librarian DB) independently identified articles eligible for further review by screening of identified titles and abstracts. Articles were considered for inclusion in the meta-analysis if they reported data from an original prospective cohort study with an assumed healthy study population of adult men and/or women at study baseline. Studies had to assess intake of one or all of IP-TFA, R-TFA and TFA from any source (total-TFA), and one of the outcomes had to be the incidence of fatal myocardial infarction, non-fatal myocardial infarction or total CHD-related deaths. Studies were included for further screening if one or both reviewers decided the study to be possibly eligible. The second screening was based on reading of the full-text versions. Any disagreement between reviewers was resolved by consensus.
The included studies were scrutinized and reviewed without blinding of reviewers. One investigator (NTB) was responsible for the extraction of data, which was then checked by another reviewer (RC). Data were extracted using a customized data extraction form. Relevant data included the first author's name, year of publication, baseline year, name of cohort, country of origin, number of participants (men and women) at baseline, participant eligibility criteria, age of participants, duration of follow-up, dietary assessment method and validity of the method, number of dietary assessments, TFA mean or median intake and extreme quantiles of intake, outcome assessment method, number of events, unit of measurement, confounders adjusted for in the statistical analysis, relative risks of CHD (from the most adjusted analysis) comparing extreme quantiles of TFA intake or per unit of TFA intake, and corresponding 95% confidence intervals (CIs).
The most adjusted risk ratio (RR) was used as a measure of the association between TFA and CHD. Summary estimates were obtained using inverse variance random-effects meta-analysis, with the DerSimonian and Laird estimate of the variance τ2 as a measure of heterogeneity between trials (Dersimonian and Laird, 1986). We applied random-effects, rather than fixed-effects models to estimate pooled RRs in order to take into account the heterogeneity, however small, of the risk estimates and thereby be more conservative. Meta-analyses were performed by using Review Manager 5.0 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark; http://www.cc-ims.net/RevMan). We assessed homogeneity of effects across studies using the Q-test and quantified by I2 index (Higgins and Thompson, 2002), which represents the percentage of total variation across studies that is attributable to inconsistency (heterogeneity) rather than chance (Higgins et al., 2003).
In accordance with the protocol, we performed secondary analyses for intake of R-TFA and IP-TFA, respectively. Sample size was inadequate for analyses stratifying for sex.
We examined the influence of an individual study on the pooled estimate of RR by excluding each study in turn. We specified sensitivity analyses with the aim to evaluate the influence of study quality on effect estimates as follows: (1) an analysis stratifying for the dietary assessment tool (diet records, validated food frequency questionnaire (FFQ), non-validated FFQ, diet history, 24-h recall) where the former was rated as better; (2) a meta-regression-analysis to examine if the number of confounders adjusted for in each study affected the effect size; and (3) a meta-regression-analysis to examine if the number of stars obtained in the Newcastle-Ottawa Scale quality assessment affected the effect size. The Newcastle-Ottawa Scale is one of the more comprehensive instruments for assessing the quality of non-randomized studies in meta-analyses (Wells et al., 2009). The eight-item instrument consists of three subscales, namely, selection of subjects (four-item), comparability of subjects (one-item) and assessment of outcome/exposure (three-item). For each item, each study got a score of zero, one or two stars, with a maximal total score of nine stars.
A total of 257 unique references were identified through the literature search and an additional 12 from reference lists, giving a total of 269. Of these, only eight met the selection criteria as shown in Figure 1. Three publications presented results from the Nurses’ Health Study (NHS) after 8, 14 and 20 years of follow-up, respectively (Willett et al., 1993; Hu et al., 1997; Oh et al., 2005). Since only the first publication (Willett et al., 1993) presented analysis of the association of IP-TFA and R-TFA intake with CHD, we extracted these data from this publication. For results on the association of total-TFA intake with CHD, we extracted data from the publication with the longest follow-up of 20 years (Oh et al., 2005). One of the selected studies, the Strong Heart Study (SHS) included subjects who were diagnosed with diabetes at baseline and adjusted for diabetes status in the statistical analyses (Xu et al., 2006). We contacted the authors of all the selected studies to request updated analysis or additional information. The author of one study was able to respond to our request (Jakobsen et al., 2008).
In addition, we contacted the principal investigators of five prospective cohort studies who had not published data on the association between TFA intake and CHD despite having collected relevant data. The principal investigators of the Västerbotten Intervention Program (Hallmans et al., 2003) and the Womens’ Health Study (Liu et al., 2002) could not accommodate our request and the reporting of TFA intake data was judged to be of inadequate quality by the principal investigators of the Atherosclerosis Risk in Communities Study (Folsom et al., 1997). The Iowa Womens’ Health Study (IWHS) performed a relevant analysis on our request (Personal Communication with Dr Kim Robien) and the principal investigators of the Finnish Mobile Clinic Health Examination Survey (FMC) provided us with the results of an unpublished analysis (Personal Communication with Dr Paul Knekt). Thereby, data from seven peer-reviewed articles were included in this review (Willett et al., 1993; Ascherio et al., 1996; Pietinen et al., 1997; Oomen et al., 2001; Oh et al., 2005; Xu et al., 2006; Jakobsen et al., 2008) as well as unpublished data from two prospective cohort studies.
The study design characteristics of the nine studies are presented in Table 1. Seven studies evaluated the association of total-TFA intake with fatal and/or total CHD (Ascherio et al., 1996; Pietinen et al., 1997; Oomen et al., 2001; Oh et al., 2005; Xu et al., 2006) (IWHS and FMC, unpublished); four examined the association of R-TFA with fatal or total CHD (Willett et al., 1993; Pietinen et al., 1997; Oomen et al., 2001; Jakobsen et al., 2008) and three considered the association of IP-TFA with fatal or total CHD (Willett et al., 1993; Pietinen et al., 1997; Oomen et al., 2001). The studies included from 667 to 78 778 participants who were followed for periods ranging from 6 to 21 years. Three studies included only men, three studied only women and three studies both men and women. The published studies’ quality was given a rating of 6 to 8 (out of 9) when assessed by the Newcastle-Ottawa Scale (Supplementary Table 1). Different methods for assessing dietary TFA were applied in the included studies: five used validated FFQs, one used 7-day weighed food records, two used the dietary history method, and one study used single 24-h recalls. Only one study performed repeated dietary assessments (Oh et al., 2005). The risk estimates for the most fully adjusted analysis for the individual studies are presented in Table 2, and the covariates adjusted for in Table 3. The number of covariates ranged from 8 to 22.
We pooled the individual studies’ risk estimates for comparison of extreme quintiles of total-TFA intake (Figure 2a) and found that total-TFA intake was associated with an increased risk of CHD events of 22 % (RR=1.22; 95% CI: 1.08, 1.38; P=0.002) and an almost similar risk of fatal CHD (RR=1.24; CI: 1.07, 1.43; P=0.003). The risk estimates for the individual studies corresponded to variable intake spans of total-TFA ranging from 2.8 g/day to 4 E% (or ∼10 g/day). Yet, there was no indication of heterogeneity between the studies; the I2-value was below 15% for both analyses. When omitting the two unpublished studies, the risk estimate for fatal CHD changed to RR=1.37 (95% CI: 1.13, 1.68; P=0.002), and when omitting the study that included subjects diagnosed with diabetes at baseline (Xu et al., 2006), the effect estimates increased modestly to 1.25 (CI: 1.09–1.45; P=0.002) and 1.26 (CI: 1.09–1.47; P=0.003) for total and fatal CHD, respectively.
There was no significant association between R-TFA intake and risk of CHD events when pooling all available estimates for risk associated with increasing R-TFA intake (one study compared extreme quintiles of intake (Willett et al., 1993), and two studies reported estimates for increments of 0.5 g/day (Jakobsen et al., 2008) or 0.5 E% (Oomen et al., 2001)) (RR=0.93; CI: 0.74–1.18; P=0.56; Figure 2b). Inclusion of one study evaluating the risk of fatal CHD (comparing extreme intake quintiles (Pietinen et al., 1997)) did not change this estimate (RR=0.92; CI: 0.76–1.11; P=0.36). The test for heterogeneity was not significant and the risk estimates for the individual studies corresponded to in R-TFA intake increments ranging from 0.5 to 1.9 g/day (Table 2).
Only three studies assessed the CHD risk associated with intake of IP-TFA. The pooled effect estimate suggested that IP-TFA intake increases the risk of CHD, although this association did not reach statistical significance (RR=1.21; CI: 0.97–1.50; P=0.09; Figure 2c). There was an indication of heterogeneity between the studies in this analysis (I2=66%; P=0.05), which was reduced when omitting the study of women only (I2=29%; P=0.23) whereby also the effect estimate changed (RR=1.09; CI: 0.98–1.22; P=0.12). The risk estimates for the individual studies corresponded to IP-TFA intake increments ranging from ∼1.3 g/day (0.5 E%) to 5.0 g/day (Table 2).
To address whether one single study carried most of the effect, we examined the influence of an individual study on the pooled RR estimate by excluding each study in turn. When the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study (AT/BC) was omitted from the pooled analysis of the risk of fatal CHD associated with intake of total-TFA, the risk estimate decreased and was no longer significant, RR=1.16 (95% CI: 0.97–1.38; P=0.10), whereas no other study had considerable influence on the pooled estimate.
When restricting the analysis to studies assessing TFA intake by means of validated FFQs or food records, the effect estimate for the association of total-TFA intake with CHD events or fatal CHD did not change considerably (RR=1.21; 95% CI: 1.07–1.37; P=0.002 and RR=1.28; 95% CI: 1.09–1.49; P=0.002 for total and fatal CHD, respectively), neither did the association of R-TFA with CHD (RR=0.88; 95% CI: 0.71–1.09; P=0.25). Due to the limited number of identified studies, the relatively low degree of heterogeneity and the fact that the studies were rated almost equally by means of the New-Castle-Ottawa Scale, we did not find the data suitable for further sensitivity analysis.
The present systematic review addresses the question of whether we today have enough available evidence from prospective cohort studies to evaluate if intake of IP-TFA and R-TFA, respectively, affects the risk of CHD. Our comprehensive literature search of published and unpublished results identified limited new data, whereby data from a total of only eight prospective cohort studies could be pooled.
In accordance with a previous analysis (Skeaff and Miller, 2009), we found that total-TFA intake is associated with an increased risk of fatal and total CHD of >20% when comparing extreme quintiles of intake. Inclusion of unpublished data from two cohort studies resulted in a slightly lower risk estimate of fatal CHD (RR=1.24), compared with that found previously (RR=1.32) (Skeaff and Miller, 2009). We did not calculate a risk estimate for CHD events associated with incremental TFA intake, since our literature search did not identify new data that would complement recent meta-analyses estimating that an increased TFA intake of 2 E% is associated with an RR of CHD events of 1.23 (1.11–1.37) based on data from NHS, the Health Professionals’ Follow-up Study, the Zutphen Elderly Study (ZES) and AT/BC (Mozaffarian et al., 2006; Skeaff and Miller, 2009).
Our pooled analyses of the risk associated with R-TFA and IP-TFA intake, respectively, were compromised by a very limited number of available studies, which indeed prohibits any firm conclusions about whether the source of TFA is important. The pooled estimates suggested that while dietary IP-TFA may increase the risk of CHD, R-TFA intake does not. Two studies showed strong positive association of total-TFA with CHD: the ZES which included elderly men with a very high TFA intake (Oomen et al., 2001) and the NHS (Oh et al., 2005). Whereas the association could be ascribed to IP-TFA intake in the NHS, the ZES did not indicate that R-TFA is less harmful than IP-TFA.
The validity of any meta-analysis highly depends on the quality of the included studies. Both the quality of the dietary assessment and the confounders adjusted for in the statistical analysis are central features. The lack of an effect in the SHS and the FMC (unpublished) could be due to the fact that blood lipids were adjusted for in the analyses from these studies. It is indeed problematic when potential metabolic effects of TFA intake are adjusted for, as the effects of intake of TFA on risk of CHD may be mediated through the effect of these intermediate factors. In the Health Professionals’ Follow-up Study (Ascherio et al., 1996), fiber intake was shown to be an important confounder eliminating an otherwise positive association between TFA intake and CHD. Therefore, it is positive that all studies but the FMC and the SHS (Xu et al., 2006) took this dietary factor into account. Whereas all studies adjusted for smoking, age, body mass index and intake of energy and alcohol, only five of the included studies adjusted for other subclasses of dietary fat (Oomen et al., 2001; Oh et al., 2005; Jakobsen et al., 2008) (IWHS and FMC, unpublished).
Of the included studies, the Nurses’ Health Study had the most powerful design with repeated dietary assessments using an FFQ, which specifically addressed the type and brand of margarine and a continuously updated food composition database specifically constructed for assessment of TFA composition of foods (Willett, 2006). In contrast, the validity of single 24-h recalls to estimate usual intakes of TFA is questionable and may explain why no association of TFA intake with CHD was found in the SHS (Xu et al., 2006).
In epidemiological studies, the assessment of intake of TFA is potentially affected by substantial random measurement error due to a number of factors, such as (1) participant recall bias; (2) insufficiently updated values in food composition tables; and (3) substantial changes in the TFA content of foods over time (L’Abbe et al., 2009). During the 1960s, margarine became viewed as the healthy alternative to butter because it was lower in saturated fatty acids (Willett, 2006) and many food manufacturers and restaurants replaced tallow and lard with TFA-based products. However, during the 1990s, the food industry in many countries made efforts to reduce the TFA content of margarines and shortenings and in recent years a drastic decline in the industrial-TFA content of most foods in Western countries has been reported (Craigh-Schmidt and Rong, 2009). These changes can only be accounted for by performing repeated assessment of dietary intake.
Also, a problem of confounding by indication may arise in dietary surveys, and this is difficult to correct for. A subject with symptoms of CHD or family history of myocardial infarction may be more aware of consuming a healthy diet, which in the 1970s meant replacing butter with margarine. Thereby, subjects at higher risk may have increased their margarine and with this their IP-TFA consumption while reducing their butter, and R-TFA, consumption.
Considering the difficulties in assessing TFA intake and the fact that very few prospective cohort studies have investigated the association between R-TFA or IP-TFA intake and CHD, data from other types of studies may add important evidence. The results from a case–control study (Ascherio et al., 1994) are in support of the results from the pooled analyses of cohort studies by indicating that IP-TFA was associated with increased risk, but only at intake levels above 3.3 g/day, whereas R-TFA was neutral (when comparing intakes up to 1.8 g/day).
Whereas IP-TFA intake has consistently been shown to adversely affect risk markers for CHD in controlled trials (Mozaffarian et al., 2006), very few intervention studies have examined the effect of R-TFA on cardiovascular risk markers. One study showed that when men consumed R-TFA in high amounts (3.7 E% or 10.2 g/day), the effect on blood lipids was comparable to the effects of IP-TFA by increasing the low-density lipoprotein-cholesterol and decreasing the high-density lipoprotein-cholesterol concentrations in plasma. However, when R-TFA was consumed in moderate amounts (1.5 E% or 4.2 g/day), the effect on blood lipids was not significantly different from that of a control diet with low TFA content (0.8 E% or 2.2 g/day from any source) (Motard-Belanger et al., 2008). Recently, an intervention study also found no difference in the effect of IP-TFA and R-TFA on blood lipids or insulin sensitivity at low intake levels (5 g/day) in overweight women (Tardy et al., 2009). A third intervention study suggested that R-TFA intake could affect women and men differently. Among women, both high-density lipoprotein-cholesterol and low-density lipoprotein-cholesterol were higher after intake of 5 E% (∼11–12 g/day) R-TFA compared with equivalent intakes of IP-TFA, whereas only minor differences were observed in men (Chardigny et al., 2008). In accordance, no significant difference between the effects of TFA from the two sources on the low-density lipoprotein/high-density lipoprotein-cholesterol ratio (P=0.37) was found in a recent quantitative review (Brouwer et al., 2010), which compiled the evidence from 29 and 6 treatments with IP-TFA and R-TFA intake, respectively.
In summary, the observational evidence suggests that in contrast to dietary IP-TFA, R-TFA intake does not affect the risk of CHD or may even be slightly protective. However, this could be ascribed to the fact that R-TFA generally is consumed (1) in much lower quantities than IP-TFA and (2) together with dairy products, which may be heart protective (Warensjo et al., 2010). Alternatively, the weak tendency for a risk reduction seen with R-TFA, may relate to the fact that vaccenic acid may be converted endogenously to the conjugated linoleic acid isomer c9,t11-18:2 (Turpeinen et al., 2002), the adipose tissue content of which has been shown to be inversely associated with myocardial infarction risk (Smit et al., 2010).
Denmark introduced legislation, effective from 1 January 2004, restricting the use of TFA to a maximum of 2% in oils and fats destined for human consumption. R-TFA was excluded from this legislation. The results of this systematic review support the notion that TFA intake is detrimental to the heart. However, the limited number of available studies prohibits any firm conclusions concerning whether the source of TFA is important. Any legislative discrimination between TFA from the two sources must therefore be the result of a pragmatic decision based on: (1) the low risk of achieving high daily intakes of R-TFA when consuming normal foods vs the risk of consuming considerable amount of IP-TFA with high intakes of certain food products such as fast food and snacks (or with hydrogenated vegetable oil in non-Western countries (Singh et al., 1996)); (2) the difficulties in removing R-TFA from natural food sources vs the achievable elimination of IP-TFA from most foods (Leth et al., 2006); and (3) the belief that R-TFA-containing foods are often otherwise healthy, whereby the consumption of these should not be restricted vs the notion that IP-TFA are nutritionally unnecessary.
We thank the investigators of Finnish Mobile Clinic Health Examination Survey who shared unpublished data with us: P Knekt, A Reunanen, R Seppänen, R Järvinen and A Aromaa, and K Robien and S Motzinger from the Iowa Womens’ Health Study for performing data analyses on our request and sharing the data with us. Arla Food Amba provided financial support for the execution of this review. The Parker Institute, Musculoskeletal Statistics Unit is sponsored by the Oak Foundation.
About this article
Supplementary Information accompanies the paper on European Journal of Clinical Nutrition website (http://www.nature.com/ejcn)