Cumulative distribution functions of multivariate (MV)-adjusted levels at examinations 5, 6, and 7 in men and women not treated with lipid drugs and without cardiovascular disease. Results are shown for levels of total cholesterol in (A) men and (B) women. To convert total cholesterol to millimoles per liter, multiply by 0.0259.
Cumulative distribution functions of multivariate (MV)-adjusted levels at examinations 5, 6, and 7 in men and women not treated with lipid drugs and without cardiovascular disease. Results are shown for high-density lipoprotein cholesterol (HDL-C) for (A) men and (B) women. To convert HDL-C to millimoles per liter, multiply by 0.0259.
Cumulative distribution functions of multivariate (MV)-adjusted levels at examinations 5, 6, and 7 in men and women not treated with lipid drugs and without cardiovascular disease. Results are shown for levels of triglycerides (TG) in (A) men and (B) women. To convert TG to millimoles per liter, multiply by 0.0113.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Ingelsson E, Massaro JM, Sutherland P, et al. Contemporary Trends in Dyslipidemia in the Framingham Heart Study. Arch Intern Med. 2009;169(3):279–286. doi:10.1001/archinternmed.2008.561
Recent cross-sectional population studies in the United States have shown an increase in obesity, a decrease in cholesterol values, but no changes in levels of high-density lipoprotein cholesterol (HDL-C) or triglycerides (TG).
Plasma total cholesterol, HDL-C, and TG levels, measured by the same methods at the 3 most recently completed examinations of Framingham Offspring Study participants (1991-2001), were compared in 1666 participants without prevalent cardiovascular disease, lipid therapy, or hormone replacement therapy (56% were men; mean ages of participants at the first and last examinations, 53 and 60 years, respectively). Changes in age- and multivariate-adjusted mean lipid levels were related to changes in body mass index (BMI).
Over the 3 examinations, comparing the findings of the earliest examination with those of the most recent examination, the mean HDL-C level was significantly increased (multivariate-adjusted means, 44.4 and 46.6 mg/dL in men; 56.9 and 60.1 mg/dL in women; P value for trend, P <.001 in both sexes), whereas levels of TG were decreased (144.5 and 134.1 mg/dL in men; 122.3 and 112.3 mg/dL in women; P value for trend, P = .004 in men and <.001 in women). Over the same time interval, BMI (calculated as weight in kilograms divided by height in meters squared) increased (27.8 and 28.5 in men; 27.0 and 27.6 in women; P value for trend, P < .001 in men and P = .001 in women). There was an inverse relationship between changes in BMI and magnitude of dyslipidemia (ie, individuals with the least increase in BMI had the most favorable changes in levels of HDL-C and TG).
During a 10-year period of recent examinations in the Framingham Heart Study there was a decrease in dyslipidemia with an increase in HDL-C levels and a decrease in levels of TG despite an overall increase in BMI.
The prevalence of obesity has increased dramatically over the past several decades in the United States and throughout the world.1-3 There is also evidence of an increasing prevalence of the metabolic syndrome,4,5 a condition closely linked to obesity and associated with a clustering of cardiovascular disease (CVD) risk factors including hypertension, glucose intolerance, and a dyslipidemia characterized by high levels of triglycerides (TG) and low levels of high-density lipoprotein cholesterol (HDL-C).6 Alone, or in the context of the metabolic syndrome, dyslipidemia is widely recognized as a cardiovascular risk factor that is associated with an increased likelihood of atherosclerosis.7
In the light of increasing obesity rates, it is surprising that the most recent report on lipid trends from the cross-sectional National Health and Nutrition Examination Survey (NHANES) reported that plasma levels of neither HDL-C nor TG showed significant changes from the examination period of 1988 through 1994 to the most recent examination period of 1999 through 2002.8 Over this same period, NHANES reported that low-density lipoprotein cholesterol (LDL-C) and total cholesterol levels declined, a trend first noted about 30 years earlier.9In contrast to the cross-sectional analysis performed by NHANES, the Framingham Heart Study, a large, community-based sample of men and women, conducts longitudinal follow-up examinations in the same individuals approximately every 4 years. In the Framingham Study cohort, the prevalence of obesity increased about 7% in men and 6% in women during the observed period,10 which corresponds well with the increase observed in non-Hispanic white participants in NHANES.2 Sequential examinations in the Framingham Study provide an opportunity to examine lipid trends in the same individuals over time and to put changes in lipid values in the context of increasing obesity.
This analysis was undertaken to coincide with the period of the last report from NHANES to examine longitudinal trends in HDL-C, TG, and total cholesterol levels in the Framingham Offspring cohort, a second-generation sample of Framingham men and women in their middle age.
The Framingham Offspring Study was initiated in 1971, and the design and participant selection criteria have been described previously.11 Participants who attended each of the fifth (1991-1994), sixth (1995-1998), and seventh (1998-2001) examination cycles (offspring 5, 6, and 7, respectively) were eligible for the present study (n = 3158). These examination cycles were selected a priori to roughly correspond to the period encompassed by the last 2 NHANES examinations (1988-2002). At each Framingham examination, the participants underwent a standardized medical history by a physician, a physical examination including blood pressure measurement and anthropometry, and blood sampling after an overnight fast. Participants were excluded from the current investigation if they had missing lipid level data (n = 56) or other missing covariates (n = 78) at any of the 3 examinations. Furthermore, we also excluded individuals with prevalent CVD (n = 403), and individuals prescribed lipid therapy (n = 458) or hormone replacement therapy (HRT) (n = 497) at any of the 3 examinations. Thus, the primary analyses were performed in a sample of 1666 individuals (929 men and 737 women). In secondary analyses, we analyzed trends of lipid concentrations in a larger sample, including individuals with prevalent CVD, lipid therapy, or hormone replacement therapy (HRT) (n = 3024 individuals [1410 men and 1614 women]). The study protocol was approved by Boston University Medical Center (Boston, Massachusetts) institutional review board, and all participants provided written informed consent.
The methods for measurements of lipid levels were the same for all 3 examinations, and all testing was performed by the Framingham Heart Study laboratory that, throughout this period, participated in both Centers for Disease Control and Prevention and College of American Pathologists programs for these lipid measurements. Venous blood samples were collected in tubes containing EDTA, and plasma was separated by centrifugation. Levels of plasma total cholesterol and TG were measured using automated enzymatic assay procedures.12 The HDL-C level was measured after dextran sulfate–magnesium precipitation.13 The LDL-C concentrations were estimated using the Friedewald formula.14 To exclude instrument drift or other assay-related changes over time as an explanation for changes in lipid levels, we repeated all lipid measurements from each of the 3 examinations in a subsample of 54 individuals prior to the start of this analysis. These individuals were selected to obtain samples across the range of values of each of the 3 lipid level measurements at the fifth examination cycle. The selection of samples was done blinded to lipid data at subsequent examinations. Remeasurements were performed using the same methods as had originally been employed when these samples were collected.
Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Cigarette smoking was defined by self-reported cigarette use in the year preceding the Heart Study examination. Alcohol intake was defined by self-reported number of drinks per week. Use of lipid-modifying drugs, antihypertensive drugs, and HRT was also defined based on self-report. Dietary information on consumption of total calories during the previous year was collected at each examination using a 126-item food frequency questionnaire.15
The lipid measures of primary interest for the trend analyses were total cholesterol and HDL-C levels and levels of TG. Owing to well-established sex differences in HDL-C levels, all analyses were performed separately in men and women.
In all analyses of lipid trends across examinations, we used generalized estimating equations to account for correlated observations. We constructed 2 sets of models: age-adjusted (adjusting for age and examination cycle), and multivariate-adjusted (adjusting for age, examination cycle, BMI, smoking, alcohol consumption, antihypertensive treatment, and total cholesterol level [in analyses of levels of HDL-C and TG] using covariates from the examination in question). In analyses of the larger sample including individuals with prevalent CVD, lipid therapy, or HRT, we performed additional adjustments for lipid therapy and HRT. The cumulative distribution functions of multivariate-adjusted levels of total cholesterol, HDL-C, and TG at the 3 examinations were plotted separately for men and women. In secondary analyses, we redid all multivariate models using waist circumference instead of BMI.
In addition to analyzing the trends in lipid levels as continuous variables (with trend tests of age- and multivariate-adjusted mean levels across examinations), we also compared the following proportions across the examinations using pooled logistic regressions: proportions of participants using lipid-lowering medication; proportions of participants with low LDL-C levels (<130 mg/dL), borderline LDL-C levels (130-160 mg/dL), and high LDL-C levels (≥160 mg/dL); proportions of participants with high levels of TG (>150 mg/dL); and proportions with low HDL-C levels (<40 mg/dL in men; <50 mg/dL in women). These comparisons were performed in age- and multivariate-adjusted analyses (adjusting for the covariates defined in the previous subsection), and the difference in proportions across examination cycles were compared by trend tests. (To convert HDL-C and LDL-C to millimoles per liter, multiply by 0.0259; to convert TG to millimoles per liter, multiply by 0.0113.)
To relate trends in lipid levels to changes in weight, we also performed age- and multivariate-adjusted analyses of the trends in BMI over the 3 examinations. All of the analyses were repeated in a larger sample including individuals with prevalent CVD, lipid therapy, or HRT at any of the 3 examinations.
To explore the conjoint changes in levels of HDL-C and TG, we calculated the age-adjusted pair-wise Spearman correlations between the change in levels of HDL-C and TG between examinations 5 and 7 (Δ HDL-C level and Δ level of TG) for men and women, separately. Also, we categorized all individuals into 4 groups by the pattern of change in levels of HDL-C and TG between examinations 5 and 7 (any increase in levels of both HDL-C and TG, any increase in HDL-C level, and any decrease in levels of TG, any decrease in HDL-C level and any increase in levels of TG, and any decrease in levels of both HDL-C and TG). Then, we calculated the age- and multivariate-adjusted mean changes between examinations 5 and 7 in lipid levels, fasting glucose level, and BMI in these 4 categories. Individuals without any change in levels of either HDL-C or TG were excluded from these analyses (69 men and 40 women). In an attempt to explain the observed lipid trends, we examined changes in dietary intake of total calories between examinations 5 and 7. Two-sided P values of P <.05 were considered statistically significant. All analyses were performed using SAS statistical software (version 9.1; SAS Institute, Cary, North Carolina).
Remeasurement of plasma cholesterol, TG, and HDL-C concentrations in a subset of participants with plasma available at all 3 examination cycles showed excellent correlations with original measurements over a broad range of values for each lipid fraction at each examination cycle. For comparison of remeasured lipids with original values from the earliest examination cycle of this series of cycles (examination 5) correlation coefficients were 0.985 for cholesterol, 0.997 for TG, and 0.948 for HDL-C, all significant at P < .001. The mean intra-assay coefficients of variation when comparing the original measurements with these remeasurements were as follows: for total cholesterol levels, 5% (examination 5), 3% (examination 6), and 3% (examination 7); for HDL-C levels, 7% (examination 5), 5% (examination 6), and 4% (examination 7); and for levels of TG, 6% (examination 5), 3% (examination 6), and 4% (examination 7). The clinical characteristics of the study sample at each of the 3 examinations are shown in Table 1.
Total cholesterol levels did not change over the 3 examination cycles in either sex (Table 2). In contrast, HDL-C levels increased significantly (P < .001) over the examination cycles, also when adjusting for potential confounders including total cholesterol levels. This was accompanied by decreases of levels of TG, which also were significant (P < .01) in age- and multivariate-adjusted models. These trends were essentially identical when using log-transformed TG in the models (data not shown).
In this same time period, BMI significantly increased, in age-adjusted as well as in multivariate-adjusted models (Table 2). Fasting glucose levels increased in men but not in women.
The cumulative distribution functions of multivariate-adjusted levels of total cholesterol, HDL-C, and TG for the 3 examinations are shown in Figures 1, 2, and 3, respectively. In general, these curves show a relatively similar-size shift in the distribution of each of these lipid fractions throughout the range of measured values from the earliest examination (Offspring 5) to the most recent (Offspring 7) with the distribution of values at the intermediate examination (Offspring 6) somewhat variable in relation to the earliest and most recent examination.
In a larger sample including individuals with prevalent CVD, lipid therapy, or HRT at any of the 3 examinations, there were also significant (P < .001) and substantial increases in HDL-C levels and decreases in levels of TG over time in both men and women, in age- and multivariate-adjusted analyses. Body mass index significantly (P < .001) increased over the same time period, and, in contrast to the primary analysis, there were significant (P < .002) decreases in total cholesterol levels in this larger sample.
Consistent with analyses of continuous variables, there was a significant decrease in both the proportions of individuals with low HDL-C levels (P < .001) and the proportions of individuals with high levels of TG (P = .045 to P < .004) in age- and multivariate-adjusted analyses in both men and women (Table 3). However, the categories of LDL-C levels did not change significantly over the 3 examinations (P > .60) (Table 3).
In the larger sample including individuals with prevalent CVD, lipid therapy, or HRT at any of the 3 examinations, the proportions of individuals with low levels of HDL-C or high levels of TG changed in the same manner as in the primary analysis. Also, in this sample the proportions of individuals prescribed lipid therapy increased significantly (P < .001) in age- and multivariate-adjusted analyses in both men and women, and the categories of LDL-C levels changed significantly (P < .001) over the 3 examination periods.
The age-adjusted Spearman correlations between change in HDL-C and change in TG (Δ HDL-C and Δ TG) from examination 5 to examination 7 were −0.35 in men and −0.31 in women (P < .001 for both correlations). When study participants were grouped into 4 categories according to the pattern of change in levels of HDL-C and TG between examinations 5 and 7, the largest category of change for both men and women consisted of participants with an increase in HDL-C levels and a decrease in levels of TG (Table 4). Furthermore, this category had the smallest increases of BMI between examinations 5 and 7 in both men and women in multivariate-adjusted analyses (P < .005 in men, P < .04 for women for a comparison of the group with a decrease in levels of TG and an increase in HDL-C levels with the 3 other groups).
A comparison of data from diet questionnaires at examinations 5 and 7 did not show any significant changes in caloric intake between the various lipid groups (P > .05 for all comparisons). In the multivariate-adjusted analysis there was a larger decrease in calorie intake in men with a decrease in levels of TG and an increase in HDL-C levels compared with an index group with both a decrease in levels of TG and HDL-C, however, this change was not significant (P = .64), and virtually no differences were apparent in women.
In secondary analyses in which we repeated all models using waist circumference instead of BMI, the trends in levels of HDL-C and TG were almost identical to those in the main analyses (data not shown).
We have found in the Framingham Offspring population a substantial reduction in recent years in plasma levels of TG coupled with an increase in HDL-C concentrations. This change in levels of TG and HDL-C was observed over the relatively brief period of the last 3 completed Framingham examinations (1991-2001) in both men and women; was observed throughout a broad range of TG and HDL-C values; was independent of use of lipid-modifying drugs or HRT; and, most notably, occurred in the presence of generally increasing levels of BMI. These results are in sharp contrast to those of recent broad cross-sectional population surveys in the United States,9,16 which have included individuals prescribed lipid therapy and which have not found any significant changes in plasma levels of TG or HDL-C.
Although overall there was an increase in HDL-C levels and a decrease in levels of TG in the Framingham population, it is apparent with the categorization of the population by patterns of change in levels of TG and HDL-C (Table 4) that there was appreciable heterogeneity in the change of these combined lipid fractions. In the largest segment of the population shown in Table 4 (39% of men and 38% of women), there was, notably, a decrease in levels of TG coupled with an increase in HDL-C levels. In this group, there was also the smallest increase in BMI. Thus, irrespective of the BMI increase in the population as a whole, the most favorable change in levels of TG and HDL-C in the Framingham Study was associated with the least increase in BMI. Consequently, although a population-wide increase in weight did occur in the Framingham Study during the period of these lipid level changes, this increase was largely in association with increases in levels of TG and/or decreases in HDL-C levels. Although we attempted to determine if a change in diet calories could be detected from food frequency questionnaires15 that might be correlated with the differences in BMI and lipid changes, no clear differences between these subgroups could be detected. However, we acknowledge that such post hoc analyses are secondary and should be considered only hypothesis generating.
To our knowledge, no community-based studies have reported a recent population-wide increase in blood HDL-C levels associated with a decrease in levels of TG. In the recent reports from the NHANES3 or Minnesota Heart Survey,16 the authors did not report significant changes of levels of HDL-C or TG in their primary analyses. The NHANES analysis,3 however, did show a slight increase in age-adjusted HDL-C levels in women (from 53.8 mg/dL in the 1976-1980 study period to 55.9 mg/dL in the 1999-2002 study period), but not in men. Furthermore, they found a significant increase in levels of TG when restricting their sample to adults aged 20 to 74 years. In the report from Minnesota Heart Survey16 there was no information regarding levels of TG, and HDL-C levels were found to be unchanged both in men and women in all age categories. In further contrast to the results of the NHANES3 and Minnesota4 cross-sectional surveys, which have reported decreasing values of total blood cholesterol concentrations in US adults without excluding individuals treated with lipid drugs, we did not find a change in total cholesterol values in our sequentially examined Framingham Offspring cohort who were not treated with lipid drugs.
We believe that the primary reason other population surveys may not have detected the kind of changes in levels of TG and HDL-C that we report is likely the result of the cross-sectional design of these other studies, compounded, perhaps, by a failure to separate subgroups of individuals by patterns of lipid change. Some of the differences in results may also be attributed to differences in laboratory methods between examinations or a failure to adjust analyses for potentially confounding other therapy. However, we also recognize that differences in results between repeated cross-sectional surveys and sequential population examinations may also be related to differences in dealing with the effects of age, which in cross-sectional analysis entails “age-matching,” and in a sequential study requires adjusting for age. We further recognize that the changes observed in sequential population examinations (like Framingham), and unlike those in cross-sectional studies (like NHANES3), could be influenced not just by the changing age of a population but also by previous examinations that, in particular, might provide an impetus for participants to adopt more favorable lifestyle measures (eg, decreased smoking or increased physical activity).
At present, we have no certain explanation for what can be viewed as a potentially beneficial recent change in blood lipid levels in a fairly large segment of the Framingham Heart Study population. The change we have observed, in which levels of TG and HDL-C are inversely related, would strongly imply that there is a physiologic explanation that is responsible for an initial reduction in levels of TG that is accompanied by a causally related increase in the HDL-C level. Blood levels of TG and HDL-C are well known to be reciprocally related and are changed principally through the action of the TG hydrolase, lipoprotein lipase. This enzyme catalyzes the hydrolysis of plasma triglyceride-rich lipoproteins that results not only in smaller triglyceride-rich particles but also in an increase in HDL-C particles that are newly created from the excess of polar lipids and apoproteins that originate from the surface of the triglyceride-rich lipoproteins.17,18 Indeed, animal studies suggest that with acute inhibition of lipoprotein lipase and a reduction in blood HDL-C levels, as much as 50% of the concentration of HDL-C in the blood may be the result of the catabolism of triglyceride-rich lipoproteins.19
We think it is possible that a decrease in plasma TG, coupled with an increase in HDL-C in the Framingham population, may be the result of more active and complete hydrolysis of triglyceride-rich lipoproteins. If so, the reason this may have occurred is not known, but NHANES diet surveys suggest recent large changes in patterns of food consumption in the United States,20 characterized by an increase in carbohydrates, a reduction in total fats, and especially a reduction in saturated fat. We think it is possible that a relative reduction in dietary saturated fat could result in plasma triglyceride-rich lipoprotein particles with both phospholipids and TG that have relatively more unsaturated acyl chains in most recent examinations than in previous examinations. Triglyceride-rich lipoproteins with more unsaturated acyl groups are likely to be more readily hydrolyzed than more saturated acylglycerides21,22 and consequently result in increased formation of HDL-C.
The strengths of our study include the large, community-based sample consisting of men and women with repeated and standardized lipid measurements, allowing analyses of intra-individual changes, the remeasurement of lipid levels to exclude assay-related changes over time, and the detailed information on potential confounders. Nevertheless, our study also has several limitations. As in any study of trends restricted to the same individuals, there may be a survivor bias that would more likely show a more favorable than unfavorable change in lipids. In addition, we cannot entirely rule out the possibility that knowledge of the lipid levels could have induced lifestyle changes or medication use in the participants that may have affected the lipid trends in this longitudinal cohort study (ie, a healthy cohort effect). Furthermore, our study sample consisted predominantly of middle-aged white individuals of European descent, limiting the generalizability of our findings to other age groups and ethnicities. We did not have uniform information about physical activity at each examination cycle, which might be relevant to changes in levels of TG and HDL-C. However, the results we have obtained were robust on adjustment for BMI, which is known to have an effect on levels of TG and HDL-C and might be expected to (inversely) reflect the extent of exercise. Finally, our dietary information was obtained from a food frequency questionnaire, estimating the kinds of foods consumed over a 1-year period. We recognize that, although we found no differences in calories consumed between various lipid groups, small changes in the categories of dietary nutrients that might influence levels of plasma lipids might not be detected by dietary (recall) histories.
In our longitudinal study of a large, community-based population during the period from 1991 to 2001, we observed decreasing dyslipidemia characterized by increasing HDL-C levels coupled with decreasing levels of TG. Over this same time period, BMI generally increased but increased less in those with the most favorable change in levels of TG and HDL-C. Our observation is not easily explained but may relate to diet changes, especially to changes in the kinds of fat that may influence the turnover of triglyceride-rich lipoproteins. Additional studies are warranted to confirm our findings and to elucidate the mechanisms underlying these potentially favorable trends.
Correspondence: Sander J. Robins, MD, Framingham Heart Study, Boston University School of Medicine, 73 Mount Wayte Ave, Ste 2, Framingham, MA 01702-5803 (firstname.lastname@example.org).
Accepted for Publication: August 5, 2008.
Financial Disclosure: None reported.
Funding/Support: This work was supported through the Swedish Heart-Lung Foundation (Dr Ingelsson) and National Institute of Health/ National Heart, Lung, and Blood Institute (NHLBI) contract N01-HC-25195.
Role of the Sponsor: The funding sources had no role in the study design, analyses, or drafting of the manuscript. The NHLBI reviews all manuscripts submitted for publication, but it was not involved in the decision to publish.
Author Contributions: Dr Robins had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Ingelsson, Levy, Vasan, and Robins. Acquisition of data: Massaro, Sutherland, Jacques, Levy, and Robins. Analysis and interpretation of data: Ingelsson, Massaro, Sutherland, Levy, D’Agostino, Vasan, and Robins. Drafting of the manuscript: Ingelsson and Robins. Critical revision of the manuscript for important intellectual content: Ingelsson, Massaro, Sutherland, Jacques, Levy, D’Agostino, Vasan, and Robins. Statistical analysis: Massaro and D’Agostino. Obtained funding: Ingelsson. Administrative, technical, and material support: Sutherland and Levy. Study supervision: Levy, D’Agostino, and Robins. Nutritional expertise: Jacques.