Plasma concentrations of trans fatty acids in persons with type 2 diabetes between September 2002 and April 2004^,,

Abstract

Background: transFatty acids (TFAs) increase cardiovascular disease risk. TFAs and polyunsaturated fatty acids (PUFAs) in the food supply may be declining with reciprocal increases in cis-monounsaturated fatty acids (MUFAs) and saturated fatty acids (SFAs).

Objectives: We sought to determine whether plasma 18-carbon TFA and PUFA concentrations might decrease over time and whether there might be reciprocal increases in plasma cis-MUFAs and SFAs.

Design: We studied 305 persons (171 women) taking part in Look AHEAD (Action for Health in Diabetes), a randomized trial of lifestyle intervention for weight loss to reduce major cardiovascular events in overweight and obese adults (aged 45–76 y) with type 2 diabetes who also participated in an ancillary study of oxidative stress. This study was a cross-sectional analysis of TFAs, cis-MUFAs, SFAs, and PUFAs measured in plasma before intervention (September 2002–April 2004).

Results: In a model that included demographic characteristics, plasma total fatty acid concentration, BMI, serum insulin, statin use, season, and longitudinal time trend (R² = 0.167, P < 0.0001), plasma TFAs decreased by 13.5%/y (95% CI: −22.7, −3.2%/y; absolute decrease 7.0 mg · L⁻¹ · y⁻¹; 95% CI: −12.5, −1.6 mg · L⁻¹ · y⁻¹; P = 0.012). This longitudinal trend was not significantly altered by further adjustment for dietary variables and physical activity. In contrast, longitudinal trends for PUFAs, cis-MUFAs, and SFAs were weak and not significant.

Conclusions: This change in plasma concentrations of TFAs is consistent with changes in fatty acid composition that food manufacturers are likely to have made to avoid declaring TFAs on food labels. Further research will be needed to determine the overall effect of these changes on cardiovascular risk. The Look AHEAD trial is registered at clinicaltrials.govas NCT00017953.

See corresponding editorial on page 665.

INTRODUCTION

trans Fatty acids (TFAs) have adverse metabolic effects and increase cardiovascular risk. In addition, they have adverse effects on lipoproteins (1, 2), and increasing evidence suggests that TFAs from partially hydrogenated oils increase inflammation (3, 4) and impair endothelial function (3, 4). Consistent with the adverse metabolic effects of TFAs, observational studies generally suggest that increased exposure to TFAs confers increased cardiovascular risk (5–10); a meta-analysis estimated the risk of ischemic heart disease to be increased by 23% for each 2% increase in TFAs as a percentage of energy (11).

The observed adverse effects of TFA suggest the importance of minimizing TFA exposure. Humans are exposed to TFAs via foods consumed, including fatty foods from ruminants that form TFAs in their first stomach, and from hydrogenated oils (12, 13). In the United States, partially hydrogenated oils are the major food sources of TFAs (12–14). Previous data showed reductions in dietary intake of TFAs between 1980–1982 and 1995–1997 (15). Between November 2002 and July 2003, Frito-Lay, a major supplier of snack foods in the United States, eliminated trans fats from its most popular snack foods (13), suggesting that TFAs in the food supply are likely to have decreased after July 2003. Reduction in TFAs in the food supply more generally is also likely because other members of the food service industry sought to reformulate their products (13, 15) to reduce or eliminate TFAs to avoid the mandatory labeling of packaged foods beginning January 2006 (17).

Information for changes in biomarkers of TFA exposure after changes in TFAs in the food supply is limited. Small studies in persons in Australia (8), Canada (18), Costa Rica (19), and the United States (20, 21) showed that biomarkers of TFA exposure, including plasma (20), erythrocyte membrane (21), human milk (18), and adipose tissue (8, 19) concentrations of TFAs can decrease after mandatory TFA labeling of foods (18, 20, 21) or reduction in TFAs in the food supply (8, 19). However, as far as the authors are aware, no information is available for either dietary intake or biomarkers of TFA exposure among persons with type 2 diabetes. This is an important gap because type 2 diabetes is associated with increased cardiovascular risk (22, 23), increased free fatty acid flux to the liver (22, 24), and a dyslipidemia that shares some features associated with high dietary TFA intake (25), including elevated plasma triglycerides, low HDL cholesterol, and small, dense LDL cholesterol (24).

In this study, we assessed whether plasma TFA concentrations in persons with type 2 diabetes declined longitudinally between 2002 and 2004 and whether changes in other plasma fatty acids were consistent with the expectation that fats in the food supply containing TFAs would be replaced by other fats higher in cis-MUFAs or SFAs and lower in PUFAs (13, 16).

SUBJECTS AND METHODS

Study design

This study was a cross-sectional analysis of baseline data collected as part of an ancillary study of the Look AHEAD (Action for Health in Diabetes) study, a randomized multicenter, controlled trial of a lifestyle intervention for weight loss in overweight or obese adults (aged 45–76 y) with type 2 diabetes that aimed to reduce major cardiovascular disease events (26, 27).

Study participants

Our study population included all persons taking part in Look AHEAD at the Houston and Baltimore Look AHEAD clinics who were randomly assigned to treatment between December 2002 and April 2004 and also participated in an ancillary study of oxidative stress for which they donated blood samples between 26 September 2002 and 29 April 2004. From among all of the Look AHEAD clinical sites, these centers were selected because they participated in a substudy that assessed physical activity by accelerometry (26). Participants who were recruited into Look AHEAD were also given the opportunity to participate in one or more of the ancillary studies of Look AHEAD approved by the Look AHEAD Ancillary Studies and Steering Committees. At the Houston and Baltimore Look AHEAD centers, Look AHEAD clinical staff recruited participants into ancillary studies in conjunction with recruitment into Look AHEAD. This analysis included 305 Look AHEAD participants recruited into an ancillary study of oxidative stress at a baseline (before intervention) Look AHEAD visit for whom baseline plasma concentrations of fatty acids could be measured. Information on age, sex, race, ethnicity, BMI, and medication use of participants was collected at screening/baseline by Look AHEAD staff by using standard Look AHEAD procedures (26). All participants provided voluntary informed consent. All procedures were conducted in accord with the ethical standards of the institutional review boards of the Baylor College of Medicine, Arizona State University, the Phoenix VA Health Care System, and for Johns Hopkins Medical University, the Western Institutional Review Board.

Timing of collection of individual data

All of the information presented below was collected either at screening or during baseline visits for Look AHEAD, before the intervention. Dietary information was collected a median of 31 d (IQR: 9–47 d) before collection of blood for fatty acid and insulin concentrations. Collection of information for physical activity ended at a median of 47 d (IQR: 30–70 d) before collection of blood samples. Persons who participated in this study at the Houston Look AHEAD clinic donated blood samples between 26 September 2002 and 29 April 2004 inclusive. Persons who participated in this study at the Baltimore Look AHEAD clinic donated blood samples between 2 December 2002 and 9 March 2004 inclusive. The numbers of persons (for both clinical sites combined) who donated blood samples in each of years 2002, 2003, and 2004 were 67, 196, and 42, respectively.

Study objectives

This study had 2 objectives. First, to determine how plasma TFAs changed between September 2002 and April 2004, an interval during which TFAs in the food supply was expected to decline. Second, to determine whether plasma PUFAs also declined and whether plasma cis-MUFAs and SFAs increased, in parallel with the expectation that the food industry would replace fats containing TFAs with fats lower in PUFAs and higher in cis-MUFAs or SFAs. In this context, we use the phrases “secular trend,” “advancing time,” or “longitudinal trend” to designate the longitudinal trend for plasma TFA concentrations, independent of variation that might occur seasonally.

Plasma and serum measurements

Plasma lipid profile, fasting glucose, and whole-blood glycosylated hemoglobin determined by using samples collected at a Look AHEAD baseline visit were measured in the Look AHEAD central laboratory (28). At a baseline visit at which fasting blood chemistry tests were collected for Look AHEAD, separate plasma and serum samples were collected by venipuncture for this study. Following Look AHEAD procedures, fasting status (at least 8 h without food or any beverage other than water) was verified by Look AHEAD staff by participant interview before collecting fasting blood samples. Local Look AHEAD staff collected and processed these samples by a standardized protocol after receiving training in this protocol by the first author. This training was reviewed and reinforced approximately annually. In brief, blood for isolation of plasma was protected from light, maintained on ice as much as possible, supplemented with 50 μmol butylated hydroxytoluene/L, and frozen at −70°C within 2 h of blood collection until measurement of fatty acid composition. Serum was frozen at −20°C within 2 h of blood collection until measurement of insulin. Serum insulin was measured by standard clinical methods by Quest Diagnostics with both an intraassay and interassay CV of <9%.

Plasma fatty acid concentrations were measured by gas liquid chromatography after preparing fatty acid methyl esters in a core laboratory at Wake Forest University School of Medicine by the same methods that laboratory used previously for serum from humans (29, 30). The individual fatty acids that we considered and how these fatty acids were classified into TFA, SFA, cis-MUFA, and PUFA classes are shown in Table 1. The fatty acids included in the TFA, SFA, cis-MUFA, and PUFA classes shown in Table 1 represent the predominant TFAs, SFAs, cis-MUFAs, and PUFAs detected in plasma of our study participants and account for (mean ± SD) 100%, 99.1 ± 0.5%, 98.9 ± 0.8%, and 99.5 ± 0.4% of all TFAs, SFAs, cis-MUFAs, and PUFAs of a carbon-chain length >14 detected in plasma of our study participants. Also shown in Table 1 are the intraassay and interassay CV for the individual fatty acids that were measured as well as the corresponding information for the sums that represent the TFA, SFA, cis-MUFA, and PUFA classes and total plasma fatty acids that we investigated. As indicated in Table 1 (footnote 2), some of the fatty acids present in low concentrations in plasma were not detected in some individuals. For the primary analyses, we considered fatty acids that were not detectable to have concentrations equivalent to the smallest fatty acid concentration that could be detected (a few nanograms in the assay, equivalent to a plasma concentration of 0.6 mg/L) divided by the square root of 2.

TABLE 1.

Characterization of the fatty acids included in the fatty acid classes reported¹

			CV
Notational name	Common name	Percent of class	Intraassay	Interassay
			%
Fatty acid/fatty acid class
trans Fatty acids
18:1n–7	trans-Vaccenic acid	—	—	—
18:1n–8	—	—	—	—
18:1n–9	9-Elaidic acid	—	—	—
18:1n–12	Petroselaidic acid	—	—	—
Sum		100.0	4.2	4.4
SFAs
16:0	Palmitic acid	75.7 ± 3.42	2.8	5.3
18:0	Stearic acid	24.3 ± 3.4	3.0	1.5
Sum			2.5	3.8
cis-MUFAs
16:1n–7	cis-9-Palmitoleic acid	8.7 ± 2.8	4.0	9.8
18:1n–73	cis-Vaccenic acid	7.5 ± 1.5	3.2	4.2
18:1n–9	Oleic acid	83.8 ± 3.2	2.6	3.6
Sum			2.6	4.2
PUFAs
18:2n–6	Linoleic acid	67.6 ± 5.1	3.5	7.8
18:3n–3	α-Linolenic acid	1.2 ± 0.7	5.3	11.7
18:3n–6	γ-Linolenic acid	1.4 ± 0.5	5.8	13.5
18:4n–33	α-Parinaric acid	0.29 ± 0.23	10.3	7.9
20:3n–63	Dihomo-γ-Linolenic acid	3.9 ± 1.0	3.0	5.2
20:4n–6	Arachidonic acid	19.0 ± 4.2	3.4	7.7
20:5n–3	Eicospentaenoic acid	1.5 ± 1.2	4.5	12.0
22:5n–33	Docosapentaenoic acid	1.2 ± 0.3	3.9	5.5
22:6n–3	Docosahexaenoic acid	3.9 ± 1.4	3.5	8.3
Sum			3.5	7.8
Total plasma fatty acids		100.0	2.9	5.7

¹Values are for n = 305 persons except as noted.

²Mean ± SD (all such values).

³cis-18:1n–7, 18:4n–3, 20:3n–6, and 22:5n–3 were not detected in 2, 38, 4, and 1 samples, respectively. These undetectable values were assumed to have a value equivalent to the lowest value of an individual fatty acid that could be detected (a few nanograms in the assay, equivalent to a concentration of 0.6 mg/L plasma) divided by the square root of 2.

Dietary intake of TFAs, total fat, SFAs, oleic acid, PUFAs, and total energy

Dietary intake of TFAs, total fatty acids, SFAs, oleic acid (the predominate cis-MUFA in the diet), PUFAs (n–3 and n–6 combined), and total energy was determined from the Look AHEAD food-frequency questionnaire (FFQ) that was completed at baseline and elicited information for food intake during the past 6 mo (31–33). This FFQ is based on a previously validated FFQ designed for a multiethnic population (32) to which meal replacement beverages and snack bars were added because these items were components of the Look AHEAD intervention (31). This method of dietary assessment was selected for use by Look AHEAD because it captures dietary intake of a broad range of food items that are consumed in different regions and by different ethnic groups (31, 32). Fatty acid intake was expressed in grams. In addition, dietary TFA intake was expressed as a percentage of energy to facilitate comparison with prior work.

Physical activity energy expenditure

Total daily energy expenditure in physical activity was determined by accelerometry at screening/baseline in those participants who also chose to participate in the accelerometry substudy of Look AHEAD (26) and was expressed in kilocalories per day.

Completeness of data availability

Among the 305 participants in this study, 229 had complete information for plasma TFAs and all covariates. Of the 76 participants with missing information, 40 were missing accelerometry data only, 25 were missing FFQ data only, 7 were missing both FFQ and accelerometry data, 3 were missing serum insulin only, and 1 was missing serum insulin and accelerometry data. Reasons for missing accelerometry data included equipment failure, refusal to participate in the accelerometry substudy, scheduling problems, and other reasons (34). Reasons for missing FFQ information included invalid results such as reporting too few foods or inconsistent reporting of types of foods on the FFQ, failure of the participant to complete the FFQ, or other reasons. A comparison of sociodemographic, physical, clinical, metabolic, dietary, and activity measures for those with and without complete data is shown in Supplemental Table 1 under “Supplemental data” in the online issue.

Statistical methods

All data were examined for normality, and transformations to natural logarithms were applied where needed to enhance normality. Student's t test and the chi-square statistic were used to evaluate differences between unadjusted means of continuous and categorical variables, respectively. In cases in which numbers of persons in individual categories were small, Fisher's exact test was used instead of the chi-square statistic. Heterogeneity of results for comparisons of categorical variables between sexes and between clinics was assessed with the Breslow-Day test. Associations between plasma measures and between plasma and dietary measures were evaluated with Pearson correlations adjusted for age and sex. Annual rates of change in plasma TFAs, SFAs, cis-MUFAs, and PUFAs were determined from a series of ANCOVA models. Plasma total fatty acid concentrations were included in all models to account for unknown and unmeasured effects (eg, recent weight loss) that might generally reduce plasma fatty acid concentrations.

Because of the high correlations between some of the fatty acid classes and total plasma fatty acids (Pearson correlation coefficients >0.894 for plasma PUFAs, cis-MUFAs, and SFAs), to reduce collinearity we modeled residuals from regression of plasma TFAs, SFAs, cis-MUFAs, and PUFAs (all transformed to natural logarithms) on total plasma fatty acid concentrations (transformed to natural logarithms). This approach is comparable to the nutrient residual approach (7) in modeling dietary information.

The first model included season of blood collection (categorical: fall, winter, spring, summer), longitudinal time trend, plasma total fatty acid concentration (transformed to natural logarithms), clinic, age, and sex. Longitudinal time and age were included in the first model (and subsequent models) as continuous variables without transformation (ie, as linear effects) because transforming these variables (logarithmic, to square root) did not improve model fit as assessed by residual plots. The variable representing longitudinal time trend was centered at the mean recruitment time so that annual percentage change was calculated for the period from 0.5 y before the mean recruitment time to 0.5 y after the mean recruitment time. Model 2 included additional adjustment for race-ethnicity (black, Hispanic, white, other), BMI, serum insulin concentration (transformed to natural logarithms), and statin use (yes or no). BMI was included in model 2 (and subsequent models) as a continuous variable without transformation (as a linear effect) because transformations (logarithmic, to square root, or as tertiles) did not improve model fit as assessed by residual plots. Insulin was included because insulin influences fatty acid metabolism, and statin use was included because statins promote LDL receptor–mediated clearance of lipoproteins, which include fatty acids in cholesterol esters, triglycerides, and other forms.

To account for the possibility that differences in plasma fatty acid concentrations in persons who donated blood samples at different times might be due to differences in dietary intake of fatty acids, model 3 included all variables included in model 2 and additionally included dietary variables. For each of plasma TFA, SFA, cis-MUFA, and PUFA residuals, model 3 included total dietary fatty acids and dietary TFAs [both as nutrient residuals (7)] and total energy (transformed to natural logarithms). For all plasma fatty acid classes, model 3 also included [as the nutrient residual (7)] dietary intake of the corresponding fatty acid as follows: model 3 for SFA residuals, dietary SFAs; model 3 for plasma cis-MUFA residuals, dietary oleic acid; model 3 for plasma PUFA residuals, dietary PUFAs. Model 3 for TFAs also included dietary SFAs so that time trends for TFAs could be assessed independent of dietary SFAs. Model 4 included all variables included in model 3 and additionally included physical activity as indicated by total activity energy expenditure (transformed to natural logarithms).

For model 1, information for all covariates was available for all study participants and the model was fitted to data for all persons. For models 2 through 4, some covariates were missing. We used multiple imputation to impute missing covariates to allow fitting models 2 through 4 to data for all study participants. We took this approach because whereas the fraction of missing information for covariates was small for individual persons, a relatively large proportion (76 of 305) of persons had some missing covariates. This approach avoids bias that might result by fitting models only to individuals with complete data. Five data sets were imputed by using the Markov chain Monte Carlo method. Models 2 through 4 were fitted to each imputed data set. Parameter estimates and variance estimates from models fitted to each of the 5 data sets were combined to derive valid summary parameter estimates and SEs. To confirm that the summary parameter estimates and SEs were stable, 3 replicate multiply imputed data were produced and the summary parameter estimates compared. Model R² and P values do not have associated SEs, and we are not aware of standard methods for combining R² and P values from models fitted to multiply imputed data sets. Therefore, we report below the range of R² and P values across the models fitted to the 5 multiply imputed data sets.

For each of models 1 through 4, we tested for effect modification by the geographically distant clinical sites by considering whether including clinic by season interaction, clinic by longitudinal time trend interaction, clinic by sex interaction, and clinic by age interaction (either individually or in combination) improved the fits of the models as indicated by the F statistic. Adequate fits of the final version of model 4 were documented by residual plots and other regression diagnostics (35).

To determine the geometric means for changes in plasma TFAs, SFAs, cis-MUFAs, and PUFAs, we back-transformed parameters obtained from fitting models to data for these plasma fatty acids on a natural logarithmic scale. Because fatty acid classes were modeled as residuals, in making these back transformations we first added the predicted value for each fatty acid class at the mean of the natural log-transformed total plasma fatty acid concentration. All data analyses were performed with SAS (version 9.2; SAS Institute). A 2-tailed P value <0.05 was considered significant.

Sensitivity analyses

As indicated above, in the primary analyses we included among the fatty acid classes only those individual fatty acids listed in Table 1 and assumed that fatty acids that were not detected had concentrations equivalent to the smallest fatty acid concentration that could be detected divided by the square root of 2. In sensitivity analyses, we considered 2 alternatives: 1) assigning fatty acids that were not detected the smallest value of the corresponding fatty acid that was detected divided by the square root of 2 and 2) assigning fatty acids that were not detected the concentration of zero.

RESULTS

Results for participant characteristics by clinical site are presented in Table 2. The study population included slightly more women than men and was predominantly of white race and non-Hispanic ethnicity. Of the 94 (31%) study participants who were of minority race-ethnicity, 23 of 24 Hispanic participants were recruited by the Baylor College of Medicine Look AHEAD site in Houston, whereas the numbers of African American participants recruited by the Baylor site and the Johns Hopkins Medical University site were similar. Study participants’ mean ± SD age was 61.1 ± 5.5 y, with 65% of participants between 56 and 65 y of age. Mean participant BMI (kg/m²), duration of diabetes, and glycosylated hemoglobin were 36.3, 6.9 y, and 7.1%, respectively, which is similar to the overall Look AHEAD study population (36); mean ± SD fasting glucose at 8.0 mmol/L ± 2.4 was ∼5% lower than in the overall Look AHEAD population. Plasma lipids and lipoproteins and blood pressure were also similar to the overall Look AHEAD population. Systolic blood pressure was similar in participants from both clinics, but diastolic blood pressure was 4 mm Hg lower in participants in Houston (P < 0.001). Daily activity energy expenditure for participants at the Baltimore clinic was ∼20% greater than for participants at the Houston clinic (P = 0.042 for data transformed to natural logarithms). Plasma TFAs were 25% higher for Houston participants compared with Baltimore participants (P = 0.006 for data transformed to natural logarithms). Plasma concentrations of total cis-MUFAs were 6.8% higher for Houston participants, which is consistent with the 13% higher consumption of dietary oleic acid for Houston participants (P = 0.005 and P = 0.022, respectively, for data transformed to natural logarithms). None of the other dietary or plasma fatty acids differed between clinics. Median dietary TFA intake was 2.4% of energy, which is substantially higher than the maximum of 1% of energy recommended by the American Heart Association (37).

TABLE 2.

Characteristics of study participants by clinical center

Characteristic	Baltimore clinic	Houston clinic	Combined	P value1
n2	120	185	305
Sex distribution [n (%)]				0.086
Men	60 (50)	74 (40)	134 (44)
Women	60 (50)	111 (60)	171 (56)
Racial distribution [n (%)]				<0.00013
White	89 (74)	122 (66)	211 (69)
African American	28 (23)	29 (16)	57 (19)
Hispanic	1 (1)	23 (12)	24 (8)
Other	2 (2)	11 (6)	13 (4)
Age (y)	61.9 ± 5.34	60.7 ± 5.5	61.1 ± 5.5	0.058
Age distribution [n (%)]				0.247
45–55 y	13 (11)	32 (17)	45 (15)
56–65 y	79 (66)	118 (64)	197 (65)
66–75 y	28 (23)	35 (19)	63 (21)
Systolic blood pressure (mm Hg)	132 ± 15	135 ± 18	133 ± 17	0.127
Diastolic blood pressure (mm Hg)	73 ± 9	69 ± 9	71 ± 9	0.0001
Weight (kg)	103 (91, 116)5	102 (90, 116)	102 (90,116)	0.689
Height (cm)	171 ± 10	169 ± 10	170 ± 10	0.023
BMI (kg/m2)	35.7 ± 6.0	36.6 ± 6.5	36.3 ± 6.3	0.249
Waist circumference (cm)	114 ± 13	116 ± 14	115 ± 14	0.344
Diabetes duration (y)	6.8 ± 5.9	7.0 ± 6.6	6.9 ± 6.4	0.800
Medications [n (%)]
Insulin	17 (14)	38 (21)	55 (18)	0.157
Oral hypoglycemic agent	81 (68)	148 (80)	229 (75)	0.014
Statin	64 (53)	95 (51)	159 (52)	0.735
Plasma and serum variables
Plasma cholesterol (mmol/L)	4.86 (4.32, 5.59)	4.81 (4.06, 548)	4.84 (4.16, 5.53)	0.309
LDL cholesterol (mmol/L)	2.84 (2.35, 3.44)	2.84 (2.25, 3.26)	2.84 (2.30, 3.34)	0.297
HDL cholesterol (mmol/L)	1.11 (0.96, 1.42)	1.14 (0.96, 1.32)	1.14 (0.96, 1.34)	0.296
non–HDL cholesterol (mmol/L)	3.67 (3.15, 4.27)	3.59 (3.03, 4.40)	3.62 (3.05, 4.29)	0.526
Plasma:HDL-cholesterol ratio	4.16 (3.40, 5.29)	4.16 (3.52, 5.07)	4.16 (3.48, 5.16)	0.771
Plasma triglycerides (mmol/L)	1.54 (1.02, 2.21)	1.75 (1.22, 2.37)	1.69 (1.10, 2.31)	0.119
Fasting serum insulin (mIU/L)	13.2 (9.0, 20.0)	14.4 (10.3, 22.1)	13.8 (9.6, 21.2)	0.386
Fasting plasma glucose (mmol/L)	7.8 ± 2.3	8.2 ± 2.4	8.0 ± 2.4	0.136
Glycosylated hemoglobin (%)	7.1 ± 1.1	7.0 ± 1.1	7.1 ± 1.1	0.724
Plasma trans fatty acids (mg/L)	43 (28, 62)	54 (38, 74)	49 (35, 69)	0.006
Plasma SFAs (mg/L)	957 (785, 1208)	969 (834, 1208)	965 (812, 1208)	0.128
Plasma cis-MUFAs (mg/L)	807 (619, 1030)	862 (701, 1090)	847 (675, 1065)	0.005
Plasma PUFAs (mg/L)	1543 (1311, 1702)	1487 (1308, 1745)	1503 (1308, 1732)	0.666
Plasma total fatty acids (mg/L)	3316 (2790, 3958)	3398 (2969, 4094)	3359 (2900, 4031)	0.107
Daily activity (kcal)	611 (407, 762)	515 (398, 682)	558 (404, 717)	0.042
Dietary variables
Dietary trans fatty acids (g/d)	4.5 (2.9, 7.0)	4.9 (3.2, 7.6)	4.7 (3.1, 7.4)	0.124
Dietary SFAs (g/d)	24.1 (16.3, 38.3)	25.8 (19.0, 41.6)	25.6 (18.1, 41.0)	0.169
Dietary oleic acid (g/d)	28.0 (19.4, 42.0)	31.7 (22.5, 48.0)	30.7 (21.4, 45.8)	0.022
Dietary PUFAs (g/d)	13.9 (9.6, 19.0)	15.4 (10.8, 21.7)	15.0 (10.4, 21.1)	0.081
Dietary total fat (g/d)	74.0 (52.5, 114.6)	83.3 (58.9, 122.6)	78.1 (55.9, 120.2)	0.116
Total energy intake (kcal/d)	1666 (1220, 2247)	1782 (1335, 2712)	1738 (1289, 2506)	0.126
Dietary trans fatty acids (% of energy)	2.4 (1.9, 2.9)	2.4 (1.9, 3.1)	2.4 (1.9, 3.0)	0.447

¹P values for difference between clinical sites. Except as noted, frequencies were compared by chi-square test; normally distributed variables shown as means ± SDs were compared by t test. Variables shown as medians (IQR) could be normalized by log transformation, and log-transformed values were compared by t test.

²Exceptions to these numbers are as follows: for whole-blood glycosylated hemoglobin, fasting plasma glucose, and plasma lipids and lipoproteins (Baltimore clinic: n = 116–118; Houston clinic: n = 164–165; total: n = 281–282); for serum insulin (Baltimore clinic: n = 120; Houston clinic: n = 181; total: n = 301); for dietary variables (Baltimore clinic: n = 95; Houston clinic: n = 178; total: n = 273); and for physical activity (Baltimore clinic: n = 101; Houston clinic: n = 156; total: n = 257).

³Fisher's exact test.

⁴Mean ± SD (all such values).

⁵Median; IQR in parentheses (all such values).

As shown in Supplemental Table 1 under “Supplemental data” in the online issue, all of the same measures are compared between the 229 persons with complete data for all plasma fatty acid models and the 76 persons with missing data for some of the covariates included in the plasma fatty acid models. The only differences between these 2 groups of individuals were a 5-mm Hg higher systolic blood pressure in those with complete data (P = 0.029, t test) and a higher representation of individuals from the Houston clinic among the group with complete data (P = 0.014, chi-square).

Scatter plots of plasma fatty acid residuals against the date of blood donation for the 4 fatty acid classes of interest are shown in Figure 1. Fatty acid residuals were determined by regressing log-transformed plasma fatty acid classes on log-transformed total plasma fatty acid concentrations, and thus show trends in concentrations of individual fatty acid classes across time of blood sampling independent of any differences in total plasma fatty acid concentrations across time. Also shown in Figure 1 are linear regression lines with 95% confidence bands and 95% prediction limits. As expected, for TFAs (upper left panel), there was a strong decreasing trend with a slope of −0.173/y (SE: 0.057; P < 0.003). Consistent with our hypothesis, trends were negative for PUFAs and positive for SFAs and cis-MUFAs; however, trends for these fatty acid classes were weaker than for TFAs and only significant for cis-MUFAs (slope: 0.0401; SE: 0.0157; P = 0.011).

FIGURE 1.

Scatter plots of fatty acid residuals plotted against blood collection date (n = 305). Fatty acid residuals were obtained by regressing log-transformed concentrations of fatty acid classes on log-transformed total plasma fatty acid concentrations (both in mg/L). Solid lines indicate linear regression slopes, gray bands indicate 95% confidence limits, and dotted lines indicate 95% prediction limits. Upper left panel: TFAs [slope (per y): −0.173; SE: 0.057; P = 0.0027]. Upper right panel: cis-MUFAs [slope (per y): 0.0401; SE: 0.0157; P = 0.011]. Lower left panel: PUFAs [slope (per y): −0.0198; SE: 0.0121: P = 0.102]. Lower right panel: SFAs [slope (per y): 0.0146; SE: 0.0087; P = 0.092]. TFA, trans fatty acid.

We investigated whether the longitudinal time trend for plasma concentrations of TFAs, SFAs, cis-MUFAs, and PUFAs differed between clinical sites by assessing interaction between clinical site and longitudinal time trend. These investigations were performed in simpler models that included as the only variables season, longitudinal time trend, age, sex, and clinic; in intermediate models that additionally included race-ethnicity and clinical variables; and in more complex models that also included dietary variables (dietary TFAs, dietary total fatty acids, total calories, and the dietary variable specific to each plasma fatty acid class) with or without physical activity energy expenditure. For those models for plasma TFA concentrations, the P values for the interaction between longitudinal trend and clinic ranged from 0.192 to 0.421. F tests comparing fits of parallel models that included and excluded interaction between clinic and longitudinal time trend provided P values of 0.359 in model 1, 0.159–0.165 in model 2, and 0.177–0.276 in model 4. In similar analyses for plasma cis-MUFAs and PUFAs, interaction between clinical site and longitudinal time trend also did not contribute significantly to the fit of the models. For plasma SFAs, P values for interaction between clinical site and longitudinal time were 0.068, 0.050, 0.082, and 0.093 for models 1, 2, 3, and 4, respectively, whereas P values for F tests comparing the fits of models with and without interaction between clinic and longitudinal time trend were 0.068 for model 1, 0.049–0.051 for model 2, 0.047–0.094 for model 3, and 0.076–0.104 for model 4. Thus, for SFAs, longitudinal time trends were determined separately for the 2 clinics. This was done by including interaction between clinic and longitudinal time in the models, because our sample size was not sufficient to allow analyses stratified by clinic.

The results of fitting models 1 through 4 to data for plasma fatty acid residuals are summarized in Table 3. After adjustment for season of blood sample, age, sex, clinic, and clinic by sex interaction, the parameter estimate for time trend for TFA log-transformed residuals was attenuated by 14% [model 1: −0.149/y (SE: 0.059) compared with −0.173/y (SE: 0.057); P = 0.0027, unadjusted results shown in Figure 1] and remained significant (P = 0.011). The parameter estimate for time trend for TFA residuals was slightly but not significantly attenuated by adjustment for race-ethnicity and clinical variables in model 2. Model 3 adjusted for dietary variables to account for the influence of possible differences in dietary intake of TFAs, SFAs, and total fat between persons studied at different times on plasma TFAs. After adjustment for these dietary variables in model 3, the parameter estimate for longitudinal time trend was identical to that for model 1 [−0.149/y (SE: 0.057) compared with −0.149/y (SE: 0.059)], and the significance of the parameter estimate for longitudinal time trend strengthened (P = 0.0099 for model 3 compared with P = 0.011 and P = 0.012 for models 1 and 2, respectively). Further adjustment for physical activity did not significantly alter the parameter estimate for longitudinal time trend for TFAs. Dietary TFA intake was positively associated with plasma TFA residuals in both model 3 (P = 0.007) and model 4 (P = 0.008). No other dietary variables were significantly associated with plasma TFA residuals in either of these models. In model 4, physical activity did not significantly contribute to explaining plasma TFA concentrations (P = 0.500). Change in TFAs in the year between 24 December 2002 and 24 December 2003 (Table 4), estimated from model 4, was −13.2% (95% CI: −22.5, −2.8%) and −6.8 mg/L (95% CI: −12.3, −1.4 mg/L).

TABLE 3.

Multivariable models fitted to residuals of log-transformed plasma fatty acid concentrations¹

Model and effect	R2	Estimate	SE	P value
trans Fatty acids
Model 1	0.1003			0.0002
Longitudinal trend (y)		−0.149	0.059	0.011
Model 2	0.1671–0.1672			<0.00012
Longitudinal trend (y)		−0.145	0.057	0.012
Model 3	0.1893–0.2034			<0.00012
Longitudinal trend (y)		−0.149	0.057	0.0099
Model 4	0.1905–0.2120			<0.00012
Longitudinal trend (y)		−0.142	0.058	0.015
MUFAs
Model 1	0.1265			<0.0001
Longitudinal trend (y)		0.026	0.016	0.092
PUFAs
Model 1	0.0919			0.0003
Longitudinal trend (y)		−0.0079	0.0123	0.521
SFAs
Model 1	0.095			0.0010
Longitudinal trend (y)
Baltimore clinic		0.031	0.016	0.059
Houston clinic		−0.006	0.011	0.594

¹Results are for models fitted to data for 305 individuals. Model 1 also includes adjustment for age (continuous), sex, clinic, season, and total plasma fatty acid concentration (transformed to natural logarithms), and for TFA and SFA only, clinic by sex interaction. For SFAs only, model 1 also includes interaction between clinic and longitudinal time trend. Model 2 includes all variables included in model 1 and additionally includes race-ethnicity (black, Hispanic, white, other), BMI (continuous), serum insulin concentration (transformed to natural logarithms), and statin use (yes or no). Model 3 for TFAs includes all variables included in model 2 and additionally includes total energy (transformed to natural logarithms), dietary total fatty acids, dietary trans fatty acids, and dietary SFAs (all as residuals of regression of natural log-transformed values on the natural logarithm of total energy). Model 4 includes all variables included in model 3 and additionally includes physical activity as indicated by total activity energy expenditure (transformed to natural logarithms). For model 1, for which data were complete, R² and P values for the model are shown. For models 2 through 4, for which missing covariates were imputed, the range of the models’ R² and P values are shown for models fitted to 5 imputed data sets.

²P < 0.0001 for models fitted to each of the 5 imputed data sets.

TABLE 4.

Changes in plasma trans fatty acid concentrations between 24 December 2002 and 24 December 2003 as estimated from multivariable models¹

	Percentage change		Absolute change (mg/L)
Model2	Estimate	95% CI	Estimate	95% CI
1	−13.9	−23.1, −3.4	−7.2	−12.8, −1.7
2	−13.5	−22.7, −3.2	−7.0	−12.5, −1.6
3	−13.8	−23.0, −3.6	−7.2	−12.6, −1.8
4	−13.2	−22.5, −2.8	−6.8	−12.3, −1.4

¹Results are for 305 individuals. The interval 24 December 2002 to 24 December 2003 is 6 mo before to 6 mo after the median date of blood collection for trans fatty acid analysis. Percentage changes and absolute changes (mg/L) are shown as geometric means.

²Model 1 adjusted for age (continuous), sex, clinic, season, total plasma fatty acid concentration (transformed to natural logarithms), and clinic by sex interaction. Model 2 includes all variables included in model 1 and additionally includes race-ethnicity (black, Hispanic, white, other), BMI (continuous), serum insulin concentration (transformed to natural logarithms), and statin use (yes or no). Model 3 includes all variables included in model 2 and additionally includes total energy (transformed to natural logarithms), dietary total fatty acids, dietary trans fatty acids, and dietary SFAs (all as residuals of regression of natural log-transformed values on the natural logarithm of total energy). Model 4 includes all variables included in model 3 and additionally includes physical activity as indicated by total activity energy expenditure (transformed to natural logarithms).

In contrast to the consistent and strong results for plasma TFA residuals, the limited multivariable adjustment in model 1 attenuated the longitudinal trend for plasma cis-MUFAs by 35% compared with the unadjusted results in Figure 1, such that it was no longer significant (P = 0.092; Table 3). Further adjustment for the additional covariates included in models 2, 3, and 4 did not uncover a significant longitudinal trend for plasma cis-MUFAs (P values for longitudinal time trend in models 2, 3, and 4 were 0.165, 0.190, and 0.229, respectively; not shown). However, plasma cis-MUFA concentration was inversely associated with dietary total fat (P = 0.006 for model 3 and P = 0.008 for model 4) and weakly positively associated with dietary oleic acid (P = 0.059 for model 3 and P = 0.064 for model 4) (data not presented in Tables 1–4). Similarly, multivariable adjustment further weakened the weak and nonsignificant longitudinal trend for plasma PUFA residuals (P = 0.521, model 1, Table 3 compared with P = 0.102, Figure 1; P = 0.729, P = 0.958, and P = 0.994 for models 2, 3, and 4, respectively; not shown). However, plasma PUFA concentrations were strongly and positively associated with dietary PUFAs (P < 0.0001 for both models 3 and 4) and inversely associated with dietary TFAs (P = 0.035 and P = 0.031 for models 3 and 4, respectively) (data not presented in Tables 1–4). Similarly to TFAs, physical activity did not significantly contribute to explaining either plasma cis-MUFA or PUFA concentrations (P = 0.502 and P = 0.694, respectively).

For plasma SFAs, the limited multivariable adjustment in model 1 with additional allowance for differing longitudinal trends for the 2 clinics identified a weakly positive longitudinal trend for the Baltimore clinic (P = 0.059) but no significant longitudinal trend for the Houston clinic (P = 0.594). After further adjustment for race-ethnicity and clinical variables in model 2, dietary variables in model 3, and physical activity in model 4 (not shown), P values for longitudinal trend in the Baltimore clinic were 0.058, 0.100, and 0.094, respectively, whereas the corresponding P values for the Houston clinic were 0.437, 0.466, and 0.566. Similarly to TFAs, cis-MUFAs, and PUFAs, physical activity did not significantly contribute to explaining plasma SFAs in model 4 (P = 0.382).

The sensitivity analyses indicated that all analyses were robust to differing assumptions concerning the concentrations of plasma fatty acids that were not detected in some individuals. Similarly, analyses of triplicate sets of 5 imputed data sets confirmed that the model parameters were stable.

Recruitment of participants into this study began in September 2002, after the age criterion for eligibility for Look AHEAD was increased from 45–74 y to 55–74 y (36). However, to assess the possibility that change in age at enrollment might have contributed to the observed decreasing secular trends for plasma TFAs, we verified that the age at enrollment into this study did not significantly change during recruitment. Median (IQR) ages of participants recruited during the first, second, third, and fourth quartiles of time for recruitment into this ancillary study were 60 (57, 64), 61 (58, 64), 60 (56, 65), and 61 (57, 65) y, respectively. Age did not significantly differ between quartiles of recruitment time (P = 0.34 by ANOVA) nor was there a significant linear trend for age with increasing quartile of recruitment time (P = 0.31), suggesting that change in age of participants recruited over time is unlikely to account for the observed time trend for plasma TFAs.

DISCUSSION

In this study we sought to determine whether plasma TFA concentrations decreased longitudinally, whether there would be reciprocal increases in plasma cis-MUFAs and SFAs, and whether plasma PUFAs declined. The principal findings of this study were that a significant secular trend was observed only for TFAs. Importantly, the secular trend for TFAs remained significant after adjustment for demographic, clinical, metabolic, and dietary variables and physical activity. In analyses that included adjustment for total plasma fatty acid concentration to account for unknown or unmeasured general effects on plasma fatty acids, we could not detect any significant effect of physical activity on plasma TFAs, cis-MUFAs, SFAs, or PUFAs.

The observed decreasing secular trend for plasma TFA concentrations is consistent with changes in the industrially formed TFA content of the food supply that could be expected during the study. For example, Frito-Lay, the major supplier of snack foods in the United States, eliminated TFAs from its most popular snack foods during the first half of this study (13). In the middle of this study (July 2003), the US Food and Drug Administration's final rule concerning TFA labeling of foods was published (17), which stimulated food producers to reformulate their products with new or different fats (13, 16) to be <0.5 g/serving in advance of the January 2006 requirement for declaring TFA content on food labels. Although not subject to mandated labeling, national restaurant chains also began replacing products containing TFAs with newer products containing fewer TFAs (13).

The finding that plasma TFA concentrations were positively associated with dietary TFAs and yet adjustment for dietary variables had little effect on estimated annual declines in plasma TFA concentrations merits comment. It is our understanding that the Nutrition Data System for Research nutrient database (version 4.01_30, 1999; Nutrition Coordinating Center), the chief source of nutrient values for the Look AHEAD FFQ (31), was not updated during this study. Thus, adjustment for dietary variables may not have accounted for secular declines in industrially formed TFAs in foods consumed by study participants as products with higher content of industrially formed TFAs were replaced by those with fewer TFAs and more SFAs and/or cis-MUFAs (13, 38).

Prior results for secular trends in exposure to TFAs are limited. Reports in Americans (7, 15, 39) showed secular trends in dietary TFAs. One study reported on dietary TFA intake later than 1997 (39), and none provided comparative information before and after the 2003 Food and Drug Administration ruling (17). In other countries (8, 18, 19), biomarkers of TFA exposure decreased after reduction in TFAs in the food supply (8, 19) or TFA labeling (18). Two prior reports for a decline in TFA exposure in Americans after 2003 included a brief report from NHANES in a small sample of non-Hispanic whites (20) that does not provide information for diabetes status and a report in a small subsample of the Framingham Offspring cohort (21) that included a small proportion (11%) of persons with diabetes. The former reported a 58% decline in total plasma TFA concentrations between 2000 and 2009 (20), whereas the later reported a 23% decrease in erythrocyte TFAs between 1999 and 2006 (21). If the decline in TFAs was linear between 2003 and the sample collection dates in 2006 (21) and 2009 (20), the annual rate of decline would be 7–10%/y, which is somewhat less than the annual decline in plasma TFA concentration we observed.

This study has limitations. We assessed differences in plasma TFA, SFA, cis-MUFA, and PUFA concentrations over time by comparing results in different persons. As in every observational study, we cannot exclude the possibility that residual confounding might have contributed to the differences in plasma TFAs that we observed longitudinally. However, the consistency of the findings between models and the plausibility of the findings in the context of the likely reduction in industrially formed TFAs in the food supply (13, 16) add credibility to our findings. Nonetheless, we cannot exclude the possibility that some factor other than reduction in industrially formed TFAs in the food supply contributed to our observations, and a longitudinal study will be needed to confirm secular trends for plasma TFAs and to clarify any trends for SFAs, cis-MUFAs, and PUFAs.

The study population also limits the generalizability of our findings. The racial, ethnic, and socioeconomic diversity of the Look AHEAD participants studied was relatively limited. Similarly, the age distribution of our study population was narrow, with 65% of participants between the ages of 56 and 65 y (inclusive). All study participants had type 2 diabetes and were overweight or obese. Further research will be needed to determine whether secular trends in plasma TFA concentrations differ according to race and ethnicity and whether such trends would be present in persons free of diabetes, of normal weight, or younger or older than our study participants.

Despite the limitations described above, the study has many strengths. This study benefitted from the Look AHEAD infrastructure and clinical data collected by Look AHEAD with the use of established and validated procedures. Although the limitations of the study population preclude generalization to healthy persons or those of differing demographic characteristics, the findings are important for several reasons. First, the prevalence of cardiovascular disease in persons with type 2 diabetes is high. Second, the consumption of TFAs adversely affects multiple cardiovascular risk factors (1–4) and increases cardiovascular events (5–10). Third, plasma TFA concentrations can serve as a biomarker of TFA intake (40). Fourth, TFA intake estimated for this study population was high at a median of 2.4% of energy, which is double the median of 1.2% of energy reported for the Nurses’ Health Study for the period 1994–2006 (39), and far greater than the maximum of 1% of energy recommended by the American Heart Association (37). Finally, the observed TFA intake would be predicted to increase cardiovascular risk by >23% (11).

In conclusion, our findings are consistent with previous work (15, 20) that showed that changes in the industrially produced TFA content of the food supply can reduce public exposure to TFAs and provide new evidence for a secular trend in plasma TFA concentrations in persons with type 2 diabetes. As industrially formed TFAs in the food supply are further reduced, we expect further reductions in population exposure to TFAs as indicated by dietary TFA intakes and plasma TFA concentrations that would be predicted to reduce population-wide cardiovascular risk. However, the overall impact of reduced dietary TFA intakes and plasma TFA concentrations will depend on the composition of the fats that replace TFAs. Observational studies in women (9) and experimental studies in nonhuman primates (41) suggest that cardiovascular benefits gained by reducing exposure to TFAs might be limited if fats containing TFAs are replaced by newer fats lower in PUFAs and enriched in SFAs and/or cis-MUFAs (13, 16, 42). Even as industrially formed TFAs in the food supply are currently lower than at the time of this study, our results remain relevant because the less expensive versions of margarines and savory snacks that are likely to be purchased by persons of limited economic means continued to have relatively high concentrations of TFAs after trans fat labeling was in effect (38, 43).