Nonfasting Triglycerides and Risk of Ischemic Stroke in the General Population FREE

The role of triglycerides in risk of ischemic stroke remains controversial.^{1 - 8} Two recent cohort studies reported a strong association between elevated levels of nonfasting, but not fasting, triglycerides and increased risk of myocardial infarction, ischemic heart disease, and death⁹ and total cardiovascular events,¹⁰ respectively. It is therefore possible that nonfasting triglyceride levels are also associated with increased risk of ischemic stroke.

Increased levels of nonfasting triglycerides indicate the presence of increased levels of remnants from chylomicrons and very low-density lipoproteins.⁹ These cholesterol-containing, triglyceride-rich lipoproteins penetrate the arterial endothelium^{11 - 12} and may get trapped within the subendothelial space,^{13 - 16} potentially leading to the development of atherosclerosis.^{17 - 18}

Triglyceride levels are usually measured after an 8- to 12-hour fast,¹⁹ thus excluding most remnant lipoproteins; however, except for a few hours before breakfast, most individuals are in the nonfasting state most of the time. Therefore, by mainly studying fasting rather than nonfasting triglyceride levels, several previous studies^{1 ,4 ,6 - 7} may have missed an association between triglycerides and ischemic stroke. Also, because former studies^{1 - 7} mainly focused on moderately elevated levels of triglycerides, an association of very high levels with risk of ischemic stroke could have gone unnoticed.

We tested the hypothesis that increased levels of nonfasting triglycerides are associated with risk of ischemic stroke in men and women in the general population. For this purpose, we studied 13 956 individuals from the Copenhagen City Heart Study with up to 31 years of follow-up, during which time 1529 developed ischemic stroke.

The prospective and cross-sectional studies were approved by Herlev Hospital and a Danish ethical committee (Nos. 100.2039/91 and 01-144/01, Copenhagen and Frederiksberg committee) and were conducted according to the Declaration of Helsinki. Participants provided written informed consent.

The Copenhagen City Heart Study is a prospective cardiovascular study of the Danish general population initiated in 1976.²⁰ We invited 19 329 white women and men of Danish descent, stratified into 5-year age groups from 20 to 80 years and older and drawn randomly from the national Danish Civil Registration System. Of those invited, 14 223 (74%) attended and 13 956 (72%) had nonfasting triglyceride levels determined using fresh plasma samples; women with triglyceride levels greater than 266 mg/dL (to convert to mmol/L, multiply by 0.0113) and men with levels greater than 310 mg/dL were referred to their general practitioners for further evaluation.

Participants underwent follow-up from baseline at the 1976-1978 examination through July 2007. Follow-up was complete; ie, we did not lose track of any individuals during the up to 31 years of follow-up. Eighty-three individuals who emigrated from Denmark during follow-up were censored at the date of emigration. The present prospective study on ischemic stroke comprised 13 956 individuals, 1529 with ischemic stroke and 12 697 without. Multiple events in the same individual were not considered in the statistical analysis, because the first event led to censoring of the individual.

We also studied cross-sectionally those participants in the Copenhagen City Heart Study who attended the 1991-1994 examination, because levels of nonfasting remnant lipoprotein cholesterol as well as a lipid profile were measured at this examination. Of the 16 563 individuals invited to this examination, 10 135 (61%) attended and 9637 (58%) had nonfasting lipid profiles measured on fresh plasma samples.

Diagnoses of cerebrovascular disease, including ischemic stroke (International Classification of Diseases, 8th Revision codes 431 through 438 and International Statistical Classification of Diseases, 10th Revision codes I61 through I69 + G45) were gathered from the national Danish Patient Registry and the national Danish Causes of Death Registry. For each person registered with cerebrovascular disease, hospital records were requested. To also include nonhospitalized patients with nonfatal ischemic stroke, participants were asked at 3 study examinations (conducted 1976-1978, 1981-1983, and 1991-1994) whether they had previously had a stroke. If affirmative, further information was obtained from that person's general practitioner. Experienced neurologists blinded to triglyceride values reviewed all potential cases.²¹

Possible stroke events (among hospitalized as well as nonhospitalized patients) were validated using the World Health Organization definition of stroke, ie, an acute disturbance of focal or global cerebral function with symptoms lasting longer than 24 hours or leading to death, with presumably no other reasons than of vascular origin.²² To distinguish among stroke subtypes—ie, infarction (ischemic stroke), intracerebral hemorrhage, and subarachnoid hemorrhage—either computed tomography or magnetic resonance imaging scan, autopsy, spinal fluid examination, or surgical description was necessary. The event was diagnosed as ischemic stroke if the scan did not visualize an infarction or hemorrhage but the person had symptoms that met the criteria of the stroke definition. The diagnosis of stroke was not applied to persons in whom a scan revealed signs of prior cerebrovascular disease but who had no history of any symptoms.

Alcohol drinkers were defined as persons consuming 4 units or more of alcohol weekly (12 g of alcohol per unit). Smokers were defined as active smokers. Hypertension was defined as use of antihypertensive medication, a systolic blood pressure of 140 mm Hg or greater, or a diastolic blood pressure of 90 mm Hg or greater. Atrial fibrillation was diagnosed from electrocardiographic recordings obtained at study examinations in 1976-1978, 1981-1983, and 1991-1994.²³ Furthermore, information on atrial fibrillation (International Classification of Diseases, Eighth Revision codes 427.93 and 427.94 and International Statistical Classification of Diseases, 10th Revision code I48.9) was gathered from the national Danish Patient Registry and the national Danish Causes of Death Registry. Women reported menopausal status and use of hormone therapy. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. Diabetes mellitus was defined as self-reported disease, use of insulin or oral hypoglycemic agents, or nonfasting plasma glucose level greater than 198 mg/dL (to convert to mmol/L, multiply by 0.0555).

Enzymatic methods (Boehringer Mannheim, Mannheim, Germany) were used on fresh samples to measure plasma levels of nonfasting triglycerides, total cholesterol, and high-density lipoprotein cholesterol (HDL-C). Levels of HDL-C were not measured at baseline but at the 1981-1983, 1991-1994, and 2001-2003 examinations. The coefficient of variation for measurement of triglycerides at the levels of 89 mg/dL and 283 mg/dL were 5% and 2%, respectively. Remnant lipoprotein cholesterol was calculated as total cholesterol minus cholesterol in high- and low-density lipoproteins. Levels of low-density lipoprotein cholesterol were calculated using the Friedewald equation if triglyceride levels were below 443 mg/dL and were measured directly if levels were 443 mg/dL or greater (Thermo Fisher Scientific, Waltham, Massachusetts).

All blood samples were drawn between 8 AM and 4 PM, and 82% of participants had eaten a meal within the last 3 hours of blood sampling. The remaining 18% had eaten their most recent meal more than 3 hours prior to blood sampling. We estimated that at most 3% of participants had eaten their most recent meal more than 8 hours prior to blood sampling, ie, were fasting.

Data were analyzed using Stata version 9.2 (StataCorp, College Station, Texas). Two-sided P < .05 was considered significant. Analyses were stratified a priori by sex, because we previously found large risk differences between men and women for myocardial infarction by levels of nonfasting triglycerides.⁹

In the prospective study, to examine the association of very high levels of nonfasting triglycerides with ischemic stroke, we preplanned stratification at each 89-mg/dL increase until the top group became too small for statistically meaningful comparison with the less than 89-mg/dL group. Thus, baseline nonfasting triglyceride levels were stratified into 6 groups: less than 89 mg/dL, 89 through 176 mg/dL, 177 through 265 mg/dL, 266 through 353 mg/dL, 354 through 442 mg/dL, and 443 mg/dL or greater. Also, age and multivariate adjustment were prespecified; in multivariate adjustment, we included known cardiovascular risk factors not by themselves highly associated with elevated levels of nonfasting triglycerides and remnant lipoprotein cholesterol. For comparison we also performed adjustment for BMI and diabetes mellitus and for HDL-C levels.

Cumulative incidence differences between strata of nonfasting triglyceride levels were determined using log-rank trend tests. Cox regression models using triglyceride levels in strata or on a continuous scale estimated hazard ratios (HRs) for ischemic stroke. Proportionality of hazards over time for nonfasting triglyceride levels was assessed by plotting −ln[−ln(survival)] vs ln(analysis time). Suspicion of nonparallel lines was further tested using Schoenfeld residuals. No major violations of the proportional hazard assumption were detected. For all survival statistics, age was the time scale using left truncation (or delayed entry), which implies that age is automatically adjusted for. Hazard ratios were adjusted for age alone and for age and other traditional cerebrovascular risk factors (total cholesterol level, alcohol consumption, smoking, hypertension, atrial fibrillation, and lipid-lowering therapy [in women, HRs were also adjusted for postmenopausal status and hormone therapy]). Additional adjustment for BMI and diabetes mellitus and for HDL-C levels was performed separately on both the age-adjusted model and the multivariate-adjusted model.

For analysis of the association between triglyceride levels (in strata and on a continuous scale) and risk of ischemic stroke, we stratified analysis on sex, age at study entry, hypertension, BMI, physical activity, and hormone therapy (in women). For each stratified analysis we tested for interactions between levels of nonfasting triglycerides on a continuous scale and the dichotomized stratifying covariate on risk of ischemic stroke. Evidence for stepwise increases in risk of ischemic stroke for increasing levels of nonfasting triglycerides was tested for by using a likelihood ratio test between models using triglyceride levels on a continuous scale and models using levels in strata of 89-mg/dL increases.

Information on baseline covariates was more than 99% complete; individuals with incomplete information on covariates were excluded from multivariate analysis. Data from the 1976-1978, 1981-1983, 1991-1994, and 2001-2003 examinations were used as time-dependent covariates for multivariate adjustments. Because HDL-C levels were not measured at the 1976-1978 examination, adjustment for HDL-C was based only on measurements from the 1981-1983, 1991-1994, and 2001-2003 examinations. This reduced the number of participants from 13 956 in the main analyses to 11 416 (82%) in analyses adjusted for levels of HDL-C.

Hazard ratios including confidence intervals (CIs) were corrected for regression dilution bias using a nonparametric method.²⁴ For this correction, we used nonfasting triglyceride values from 6709 individuals without lipid-lowering therapy attending both the baseline 1976-1978 examination and the 1991-1994 examination; however, the main analyses were conducted on 13 956 individuals. These 2 measurements were 15 years apart, equivalent to roughly halfway through the observation period, the ideal time difference for this correction.²⁴ A regression dilution ratio of 0.57 was computed for women and of 0.60 for men. Correction for regression dilution bias increases the effect size for risk estimates and increases the range of CIs but does not change significance levels.

In the cross-sectional study among all 9637 participants attending the 1991-1994 examination approximately 15 years after the first examination, we compared levels of nonfasting triglycerides and lipoprotein cholesterol in those who developed and did not develop (controls) ischemic stroke between the examinations. We used general linear models adjusting for total cholesterol level, alcohol consumption, smoking, hypertension, atrial fibrillation, and lipid-lowering therapy; in women, models also adjusted for postmenopausal status and hormone therapy. Participants receiving lipid-lowering therapy were excluded from analysis.

For levels of nonfasting triglycerides on a continuous scale in the prospective study with 13 956 participants and 1529 ischemic stroke events, we had 90% statistical power at a 2-sided P < .05 to detect an HR of 1.07 per 89-mg/dL increase in both men and women.

Baseline characteristics of individuals from the general population of the Copenhagen City Heart Study are shown in Table 1. The study included 13 956 individuals (6375 men, 7581 women) aged 20 to 93 years with up to 31 years of follow-up; of these, 1529 (779 men, 750 women) developed ischemic stroke. At the 1976-1978, 1981-1983, 1991-1994, and 2001-2003 examinations, 0%, 0%, 1%, and 2%, respectively, of the participants were receiving lipid-lowering therapy.

In both sexes, the cumulative incidence of ischemic stroke increased with increasing levels of nonfasting triglycerides (P < .001 by log-rank tests for trend). Men with elevated nonfasting triglyceride levels had age-adjusted HRs for ischemic stroke ranging from 1.4 (95% CI, 0.9-2.1) for triglyceride levels of 89 through 176 mg/dL to 3.2 (95% CI, 1.7-6.2) for levels of 443 mg/dL or greater vs men with nonfasting triglyceride levels of less than 89 mg/dL (P < .001 for trend) (Figure 1). Corresponding values for women ranged from 1.3 (95% CI, 1.0-1.8) for triglyceride levels of 89 through 176 mg/dL to 5.1 (95% CI, 1.7-14.8) for levels of 443 mg/dL or greater vs women with nonfasting triglyceride levels of less than 89 mg/dL (P < .001 for trend).

After multivariate adjustment, corresponding HRs (95% CIs) ranged from 1.3 (0.8-1.9) to 2.5 (1.3-4.8) in men (P < .001 for trend) and 1.3 (0.9-1.7) to 3.8 (1.3-11.1) in women (P < .001 for trend). With adjustment for systolic or diastolic blood pressures rather than the dichotomized variable hypertension, the results were similar. Additional adjustment for BMI and diabetes mellitus attenuated risk estimates in both the age- and multivariate-adjusted models (Figure 1). Additional adjustment for HDL-C levels likewise attenuated risk estimates in both the age- and multivariate-adjusted models; however, nonfasting triglyceride levels were still associated with risk of ischemic stroke (Figure 1).

The HR for ischemic stroke for each 89-mg/dL increase in nonfasting triglyceride levels was 1.24 (95% CI, 1.19-1.29; 1529 events; 52 events/10 000 person-years) after age adjustment and 1.15 (95% CI, 1.09-1.22) after multivariate adjustment (Table 2). After stratifying for sex, age-adjusted HRs for ischemic stroke were 1.14 (95% CI, 1.07-1.21; 779 events; 44 events/10 000 person-years) for men and 1.33 (95% CI, 1.20-1.48; 750 events; 64 events/10 000 person-years) for women; corresponding multivariate-adjusted HRs (95% CIs) were 1.12 (1.04-1.20) and 1.28 (1.15-1.43). There was statistical evidence for interaction between nonfasting triglyceride levels and sex on risk of ischemic stroke (age-adjusted P = .01; multivariate-adjusted P = .03), as anticipated by our a priori stratification. We found no evidence for nonlinearity between increasing levels of triglycerides and increasing risk of ischemic stroke in women (P = .99) or in men (P = .14).

No consistent evidence for statistical interaction on risk of ischemic stroke was found between nonfasting triglyceride levels and age at study entry (<55 years vs ≥55 years), hypertension (normotensive vs hypertensive), BMI (<25 vs ≥25), physical activity (active vs inactive), and hormone therapy in women (no vs yes) (Table 2). Hazard ratios for ischemic stroke for increasing strata of nonfasting triglyceride levels stratified for age at study entry, hypertension, BMI, and physical activity are shown in Figure 2 and Figure 3. Increasing levels of nonfasting triglycerides were associated with increasing risk of ischemic stroke in all subgroups examined (Table 2 and Figures 2 and 3).

In both sexes, absolute 10-year risk of ischemic stroke increased with increasing levels of nonfasting triglycerides and with increasing age (Table 3). Absolute 10-year risk of ischemic stroke ranged from 2.6% in men younger than 55 years with nonfasting triglyceride levels less than 89 mg/dL to 16.7% in men 55 years or older with triglyceride levels of 443 mg/dL or greater. Corresponding values in women were 1.9% and 12.2%.

Among individuals who participated in the 1991-1994 examination of the Copenhagen City Heart Study, men with a previous ischemic stroke vs controls had nonfasting triglyceride levels of 191 (interquartile range [IQR], 131-259) mg/dL vs 148 (IQR, 104-214) mg/dL (P < .01); corresponding values for women were 167 (IQR, 121-229) mg/dL vs 127 (IQR, 91-181) mg/dL (P < .05) (Figure 4). For remnant cholesterol, corresponding values were 38 (IQR, 26-51) mg/dL vs 29 (IQR, 20-42) mg/dL in men (P < .01) and 33 (IQR, 24-45) mg/dL vs 25 (IQR, 18-35) mg/dL in women (P < .05). In contrast, levels of low-density lipoprotein cholesterol and HDL-C did not differ between these groups.

By using levels of nonfasting rather than fasting triglycerides^{1 ,4 ,6 - 7} and by having more statistical power than any previous study,^{1 - 8} we detected a previously unnoticed association between linear increases in levels of nonfasting triglycerides and stepwise increases in risk of ischemic stroke, with no threshold effect. The notion of linear increases is supported both by the risk estimates for levels of triglycerides on a continuous scale and by the direct test for nonlinearity using increasing 89-mg/dL strata of triglyceride levels. The highest levels of nonfasting triglycerides (≥443 mg/dL) were associated with a 3- and 4-fold risk of ischemic stroke in 4% of men and 1% of women, respectively, in the general population. Even the most recent European and North American guidelines on stroke prevention do not recognize elevated triglyceride levels as a risk factor for stroke.^{8 ,25}

Mechanistically, the explanation for these findings is straightforward. Elevated levels of nonfasting triglycerides mark elevated levels of remnant from chylomicrons and very low-density lipoproteins that likely promote atherosclerosis.^{17 - 18} Because all human cells can degrade triglycerides but not cholesterol, it may not be the triglyceride content of remnants that cause atherosclerosis but rather the cholesterol content of these particles. Remnant lipoproteins can penetrate into the arterial intima^{11 - 12} and may preferentially get trapped within the subendothelial space.^{13 - 16} Since cholesterol in remnant particles cannot be degraded when taken up by intimal macrophages, these cells are transformed into cholesterol-laden foam cells, leading to fatty streak formation and eventually the development of atherosclerosis, myocardial infarction, ischemic heart disease,^{9 - 10} and—as demonstrated in the present study—ischemic stroke.

Some former studies have overadjusted for potential confounders, ie, have adjusted for covariates that themselves are associated with elevated levels of nonfasting triglycerides and remnant lipoproteins. The classic example is adjustment for levels of HDL-C: when levels of triglycerides and remnant lipoproteins are elevated, levels of HDL-C are reduced—and vice versa for low levels of triglycerides. As a consequence, many researchers and even several large pharmaceutical companies have focused research solely on HDL-C levels and the so-called “reverse cholesterol transport” and have largely ignored levels of triglycerides and remnant lipoproteins. The strong association of HDL-C with risk of cardiovascular disease may be explained in part by the association between elevated levels of nonfasting triglycerides or remnant cholesterol and cardiovascular disease.⁹ In support of this idea, we recently observed that loss-of-function mutations in the adenosine triphosphate–binding cassette transporter 1 gene are associated with low HDL-C levels without elevated levels of nonfasting triglycerides or remnant cholesterol but are not associated with the expected increased risk of ischemic heart disease.²⁶ Furthermore, torcetrapib, a cholesteryl ester transfer protein inhibitor that increases plasma levels of HDL-C, failed to protect against the progression of atherosclerosis and even increased the risk of cardiovascular disease, cardiovascular mortality, and overall mortality.²⁷ Likewise, adjustment for diabetes mellitus and overweight also tends to mask the association between elevated levels of nonfasting triglycerides and cardiovascular disease.⁹ This is because elevated levels of remnants may contribute to the increased cardiovascular risk observed in overweight individuals, those with diabetes mellitus, or both.¹⁸ In accordance with this, adjustment for age, BMI, and diabetes mellitus or for age and HDL-C level slightly attenuated risk estimates for ischemic stroke for increasing levels of nonfasting triglycerides.

Patients with familial forms of hypertriglyceridemia have increased risk of cardiovascular death,²⁸ while patients with remnant hyperlipidemia have premature atherosclerosis.²⁹ In accordance with this, those with ischemic stroke in the Copenhagen City Heart Study who attended the 1991-1994 examination had elevated levels of nonfasting triglycerides and remnant cholesterol. Also, heterozygosity for genetic defects in lipoprotein lipase, the plasma enzyme that degrades triglycerides, is associated with elevated triglyceride levels as well as with increased risk of ischemic heart disease.^{30 - 32} Furthermore, subanalyses of 3 randomized double-blind trials suggest that among patients with elevated triglyceride levels, a 20% to 40% reduction in triglyceride levels achieved using fibrates was associated with a 30% to 40% reduction in risk of ischemic heart disease.^{33 - 35} Finally, in the Coronary Drug Project, a 26% reduction in triglyceride levels achieved using niacin was associated with a 24% reduction in cerebrovascular events.³⁶

An association between elevated triglyceride levels and risk of cardiovascular disease has previously been questioned, since extremely high levels of triglycerides (>2200 mg/dL), as seen in familial lipoprotein lipase deficiency with the chylomicronemia syndrome, does not lead to accelerated atherosclerosis.³⁰ The simple and straightforward explanation for this apparent paradox is that at such extreme triglyceride levels, lipoproteins are very large³⁷ and consequently not able to penetrate the intima of arteries.³⁸ In support of this explanation, patients with lipoprotein lipase deficiency who, during part of their lives because of lipid-lowering treatment, have triglyceride levels as low as 266 to 616 mg/dL—and thus much smaller triglyceride-rich lipoproteins—do develop premature atherosclerosis.³⁹

Another argument often presented against using nonfasting triglyceride levels for assessment of cardiovascular risk is that triglyceride levels vary greatly after intake of fatty meals. However, this is mainly based on information from fat-tolerance tests in which participants ingest 1 g of fat per kilogram of body weight and in which triglycerides may increase up to a level of 176 to 353 mg/dL 4 hours after fat intake.^{9 ,40} In contrast, triglyceride levels only increased to 140 mg/dL 4 hours after normal intake of food in individuals in the general population.⁹

The potential difficulty in using nonfasting triglyceride levels in clinical practice without standardization of the time since the most recent meal and the content of that meal should also be considered. First, because lipid profiles usually are measured in fasting individuals, nonfasting levels would require yet another blood sampling, unless nonfasting lipid levels were to become the standard in the future. Second, our results show that even random levels of triglycerides are associated with ischemic stroke; however, whether standardization of the time since the most recent meal and the content of that meal would further improve the association between nonfasting triglyceride levels and risk of ischemic stroke needs to be studied.

In support of our findings, the Women's Health Study showed that elevated nonfasting triglyceride levels, but not fasting levels, are associated with increased risk of ischemic stroke.¹⁰ Another study also found an association between elevated levels of nonfasting triglycerides and increased risk of ischemic stroke,³ while 2 others found no association.^{2 ,5} Regarding fasting triglycerides, some inconsistency exists; for example, a large Asian-Pacific meta-analysis with fasting lipid measurements in 90% of participants reported a positive association with risk of ischemic stroke,⁷ while other studies found no association.^{1 ,4 ,6 ,10} None of these former studies on fasting as well as on nonfasting triglyceride levels attempted to associate very high levels of triglycerides with risk of ischemic stroke and therefore were not able to detect our result that nonfasting triglyceride levels of 443 mg/dL or greater are associated with a 3- and 4-fold increased risk of ischemic stroke in men and women, respectively.

Limitations include that we only studied white individuals, and therefore our results may not necessarily apply to other racial groups. Also, because we did not measure levels of remnant lipoproteins at baseline, we cannot implicate these lipoproteins directly in risk of ischemic stroke; however, the finding at the 1991-1994 examination that participants with previous ischemic stroke had elevated levels of nonfasting triglycerides and remnant lipoproteins supports this hypothesis. Moreover, because 82% of our participants ate within 3 hours of the blood sampling and at least 97% ate within 8 hours of sampling, we could not compare fasting and nonfasting triglyceride levels for association with ischemic stroke. This does not diminish the finding that nonfasting triglyceride levels are associated with ischemic stroke, but it makes it difficult to know if fundamental differences exist between the fasting and nonfasting states. However, results from the Women's Health Study suggested that levels of nonfasting, but not fasting, triglycerides are associated with increased risk of ischemic stroke.¹⁰

Our study has several strengths. We studied a homogeneous white general population of 100% Danes; we had up to 31 years of complete follow-up; end points were validated using records from hospitals, general practitioners, or both; nonfasting triglyceride levels were measured at baseline in 1976-1978 using fresh samples; less than 1% of study participants were given lipid-lowering therapy; we corrected for regression dilution bias; and we had sufficient statistical power to examine even the association of very high levels of triglycerides with ischemic stroke in men as well as women.

Our results, together with those from 2 previous studies,^{9 - 10} suggest that elevated levels of nonfasting triglycerides and remnant lipoprotein cholesterol could be considered together with elevated levels of low-density lipoprotein cholesterol for prediction of cardiovascular risk. However, these findings require replication in other populations.