ORIGINAL INVESTIGATION. Relation of Triglyceride Levels, Fasting and Nonfasting, to Fatal and Nonfatal Coronary Heart Disease

ORIGINAL INVESTIGATION Relation of Triglyceride Levels, Fasting and Nonfasting, to Fatal and Nonfatal Coronary Heart Disease Lynn E. Eberly, PhD; Jeremiah Stamler, MD; James D. Neaton, PhD; for the Multiple Risk Factor Intervention Trial Research Group Background: It remains unclear whether hypertriglyceridemia is an independent risk factor for coronary heart disease (CHD), and whether fasting and nonfasting triglyceride (TG) levels are equally predictive. Methods: A total of 289 (of 12866) men randomized during 1973 through 1975 into the Multiple Risk Factor Intervention Trial with fasting and nonfasting TG levels measured at baseline were followed up for CHD incidence and death. Proportional hazards regression models were used to assess associations of fasting and nonfasting TG levels with CHD. Results: Average fasting and nonfasting TG levels were 187 and 284 (2.11 and 3.21 mmol/l), respectively. Prevalence of hypertriglyceridemia (2 [2.26 mmol/l] or more) was 31% for fasting and 61% for nonfasting. There were 175 nonfatal or fatal CHD events during 8 years and 328 CHD deaths during 25 years. Compared with TG levels less than 2, risk factor adjusted hazard ratios for CHD mortality for hypertriglyceridemia were 1.24 (P=.9) for fasting and 1.26 (P=.7) for nonfasting. For nonfatal or fatal CHD, fasting and nonfasting TG levels were similarly predictive with hazard ratios of 1.64 (P=.4) for fasting and 1.46 (P=.3) for nonfasting. These associations for fasting TG levels were assessed to be underestimated by 56% because of regression dilution bias, with attenuation likely greater for nonfasting TG levels. Conclusions: Greater ease of obtaining nonfasting than fasting measurements, greater prevalence of hypertriglyceridemia with nonfasting than fasting values, and similarly increased risk with each indicate that nonfasting TG levels may be more useful than fasting ones for risk stratification. Arch Intern Med. 23;163:177-183 From the Division of Biostatistics, School of Public Health, University of Minnesota, Minneapolis (Drs Eberly and Neaton); and Department of Preventive Medicine, Northwestern University Medical School, Chicago, Ill (Dr Stamler). The principal investigators and senior staff of the Multiple Risk Factor Intervention Trial clinical, coordinating, and support centers and the National Heart, Lung, and Blood Institute project office were published previously (JAMA. 1982;248:1476-1477). The authors have no relevant financial interest in this article. THE ROLE OF blood triglyceride (TG) levels in predicting coronary heart disease (CHD) independently of established major risk factors (total cholesterol level, low-density lipoprotein cholesterol (LDL-C) level, blood pressure, and smoking) remains unclear. In addition, it remains to be determined whether fasting or nonfasting levels are more informative for CHD risk. A National Institutes of Health Consensus Conference in 1993 on these issues concluded the following: TG levels are weak predictors at best for CHD when other risk factors are taken into account, and, on the basis of scant available data, nonfasting TG levels may be more important than fasting TG levels. 1 Research since then has not further clarified these issues. Assessment of any clinical association of TG levels with CHD is difficult because of high intraindividual variability in TG levels, which can result in sizable regression dilution, 2 skewness of population TG distributions, and high correlation of TG levels with those of other lipids, especially the inverse correlation with high-density lipoprotein cholesterol (HDL-C) levels. Although nearly all studies have shown that TG levels are a significant univariate predictor of CHD, only a few have shown that TG levels are predictive of CHD after adjustment for LDL-C and HDL-C levels and other risk factors. The Adult Treatment Panel III 3 recently recommended treatment strategies for CHD prevention for people with high (2-499 [2.26-5.63 mmol/ L]) or very high ( 5 [ 5.64 mmol/l]) fasting TG levels. Previous work on the Multiple Risk Factor Intervention Trial (MRFIT) data 4 used the control (no intervention) group to examine whether fasting TG levels are a significant predictor of fatal or nonfatal myocardial infarction independent of the ratio of fasting total cholesterol to HDL-C level. Since no study has compared the predictive values of fasting vs nonfasting TG 177

levels in the same cohort, the National Institutes of Health Consensus Panel question on their comparative value in assessing CHD risk remains unanswered. This work explores that question and quantifies the association of fasting and nonfasting TG levels with incidence of nonfatal plus fatal CHD and with CHD mortality after adjustment for major CHD risk factors and regression dilution. METHODS STUDY COHORT The design and methods of the MRFIT have been detailed. 5-7 Briefly, the MRFIT was a randomized primary prevention trial of CHD mortality among men aged 35 to 57 years at increased risk but without clinical evidence of definite CHD at baseline. Intervention consisted of dietary advice to lower blood cholesterol level, smoking cessation counseling, and stepped-care hypertension medication. During 1973 through 1975, 22 clinical centers in 18 US cities screened 361662 men for eligibility. 7 After 3 screening visits, 12866 men were randomized into the trial. Of these, 2863 (from 17 clinical centers) had TG levels measured along with total serum cholesterol from a serum sample taken at the initial screen (screen 1) with no fasting required; we refer to this as the nonfasting TG value, even though time since the last meal was not recorded. At screen 2, all participants had TG levels and other lipid fractions measured from a fasting ( 12 hours) plasma sample. This study concerns the 2863 men with both nonfasting and fasting TG levels measured at screens 1 and 2, respectively. RISK FACTOR DATA Both fasting and nonfasting TG levels were determined by the central MRFIT laboratory. Nonfasting TG level was determined from a serum sample taken at screen 1 with no fasting requirement. Fasting total cholesterol, TG, HDL-C, and LDL-C levels were determined from a plasma sample taken at screen 2, 8 on average 44 days after screen 1. Serum levels of uric acid and glucose were also determined at screen 2. Systolic blood pressure is defined as the average of 4 random-zero manometer readings, ie, readings 2 and 3 at each of screen 2 and screen 3 (the randomization visit, on average 19 days after screen 2). Height (inches), weight (pounds), self-reported alcohol use (average number of alcoholic drinks per week), years of education, marital status, working status, and approximate total family income were recorded at screen 3. Race and self-reported current smoking level (average number of cigarettes smoked per day) were recorded at screen 1. FOLLOW-UP DATA Baseline for computing time to CHD death was the randomization date. Mortality during the active intervention phase of MRFIT (through February 28, 1982) was verified by clinical staff and coded by means of the International Classification of Diseases, Ninth Revision (ICD-9). 7,9 Posttrial mortality through December 31, 199, was determined by matching identifying information, provided by each participant during screening, with National Death Index records. 1-12 Death certificates were obtained to ascertain underlying cause of death and were coded independently by 2 nosologists; a third nosologist adjudicated any disagreements. Death dates and corresponding ICD-9 or ICD-1 13 causes from January 1, 1991, through December 31, 1999, were obtained by means of the National Death Index Plus service. Coronary heart disease death was defined by ICD-9 codes 41 to 414 or 429.2 and by ICD-1 codes I-2 to I-25. Nonfatal CHD was defined as either a clinical myocardial infarction or a significant serial electrocardiogram change indicative of myocardial infarction, 14 with follow-up through trial end (February 28, 1982). STATISTICAL ANALYSES Baseline characteristics were summarized separately for those with high or very high TG level ( 2 [2.26 mmol/l]) on the basis of Adult Treatment Panel III guidelines 3 vs those with lower levels. Univariate regression analysis of each of fasting and nonfasting TG levels on each characteristic was used to examine correlates of TG levels. Cumulative event probability curves for time to CHD death and for time to nonfatal or fatal CHD were calculated for those with TG level of 2 or more and compared with those with TG levels less than 2 for fasting and for nonfasting levels separately. Univariate and multivariate proportional hazards regression models 15 were carried out for CHD death, with stratification by clinical center. Triglyceride levels were examined in separate models for fasting and nonfasting. Triglyceride levels were highly skewed to the right; therefore, models considered both natural log transformed TG levels and dichotomized levels: 2 or more vs less than 2. Potential confounders considered in multivariate models were randomized treatment group, age, body mass index (calculated as weight in kilograms divided by the square of height in meters), race (African American or not), education (years), marital status (married or not), family income, part-time worker, alcoholic drinks per week, cigarettes smoked per day, systolic blood pressure, and fasting HDL-C, LDL-C, glucose, and uric acid levels; prespecified interactions were tested. We verified the proportional hazards regression assumption for each predictor with the test of Grambsch and Therneau. 16 Similar proportional hazards models were considered for the combined nonfatal plus fatal CHD end point. The P values shown are 2-tailed, without adjustment for multiple comparisons. Effects of regression dilution were assessed on the basis of data for all MRFIT participants who had annual visit 2 fasting TG levels and baseline risk factor values (N=11634). Following the methods of Clarke et al, 2 we derived a nonparametric regression dilution factor by (1) grouping all observations into screen 2 fasting TG quintiles; (2) computing within-group means of screen 2 fasting TGs and withingroup means of annual visit 2 fasting TG levels; and (3) computing ratios of group-group differences in these means. A parametric regression dilution factor was computed as the inverse of the correlation between the screen 2 and annual visit 2 fasting TG levels. We then estimated the true TG CHD mortality association by inflating our TG regression coefficients by each of the estimated regression dilution factors. RESULTS BASELINE FINDINGS Of the 2863 men with both TG values, 54 had missing baseline data (46 for glucose, an additional 6 for HDL-C, and 2 for alcohol use) and were excluded. The remaining 289 are the basis for our analysis. Mean TG levels were 186.7 (2.11 mmol/l) (SD, 133.5 [1.51 mmol/l]) for fasting and 284.3 (3.21 mmol/l) (SD, 193.3 [2.18 mmol/l]) for nonfasting levels. Of the 289 men, 874 (31%) had fasting TG levels of 2 or more and 1724 (61%) had nonfasting TG levels of 2 or more. The percentages with TG level of 5 178

Table 1. Baseline Characteristics of 289 MRFIT Participants With Both a Baseline Fasting and Nonfasting Triglyceride Measure, Overall and by Triglyceride Level Category* Fasting TG Nonfasting TG 2 2 2 2 Overall Age, y 46.6 ± 6. 45.6 ± 6. 46.2 ± 6.2 46.3 ± 6. 46.3 ± 6. Race, % African American 8 3 9 5 6 BMI 27.4 ± 3.4 28.4 ± 3.5 27.2 ± 3.4 28. ± 3.4 27.7 ± 3.4 Serum nonfasting TG, 221.2 ± 12.7 424.1 ± 246.4 138.7 ± 37.5 376. ± 197.2 284.3 ± 194.3 Serum nonfasting TG 2, % 49 89 1 61 Plasma fasting TG, 125.9 ± 39.1 321.2 ± 166.2 124.4 ± 55.4 225.9 ± 152.1 186.7 ± 133.5 Serum total nonfasting cholesterol, 248.5 ± 34.3 261. ± 34. 243.8 ± 34.2 257.7 ± 34. 252.3 ± 34.7 Plasma total fasting cholesterol, 232.9 ± 33.8 248.9 ± 35.5 231.4 ± 33.3 242. ± 35.6 237.9 ± 35.1 Plasma fasting HDL-C, 44.9 ± 11. 36.4 ± 9.4 46.5 ± 11.6 39.6 ± 1.1 42.2 ± 11.2 Plasma fasting LDL-C, 162.9 ± 32.5 149.3 ± 37. 16.1 ± 31.9 157.8 ± 36.1 158.7 ± 34.5 Plasma fasting glucose, 99. ± 13.7 12.1 ± 17.3 98.7 ± 12.6 1.7 ± 16.3 1. ± 15. SBP, mm Hg 134.6 ± 14.3 133.8 ± 14.3 134.7 ± 14.2 134.2 ± 14.3 134.4 ± 14.3 DBP, mm Hg 89.4 ± 9.1 89.7 ± 9. 89.4 ± 8.9 89.6 ± 9.2 89.5 ± 9.1 Uric acid, 6.5 ± 1.2 7.1 ± 1.4 6.5 ± 1.2 6.9 ± 1.3 6.7 ± 1.3 Drinkers of alcohol, % 92 94 92 93 93 Alcohol use among drinkers: drinks/wk 12.3 ± 11. 13.8 ± 12.9 13. ± 11.4 12.7 ± 11.8 12.8 ± 11.7 Cigarette smokers, % 57 61 59 57 58 Cigarette use among smokers: cigarettes/d 32.7 ± 14.8 34.9 ± 14.7 33.1 ± 14.9 33.6 ± 14.7 33.4 ± 14.8 Education, y 14.1 ± 2.9 13.9 ± 3. 14.2 ± 3. 13.9 ± 2.9 14. ± 2.9 Family income, $ 21 27 ± 9352 21 661 ± 9327 21 64 ± 9383 21 189 ± 932 21 349 ± 9345 Marital status, % married 89 91 88 91 9 Employment status, % part-time 5 4 5 5 5 No. of participants 1935 874 185 1724 289 CHD deaths during 25 y, No. (%) 211 (11) 117 (13) 12 (9) 226 (13) 328 (12) Fatal or nonfatal CHD events during 8 y, No. (%) 12 (5) 73 (8) 5 (5) 125 (7) 175 (6) Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by the square of height in meters); CHD, coronary heart disease; DBP, diastolic blood pressure; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; MRFIT, Multiple Risk Factor Intervention Trial; SBP, systolic blood pressure; TG, triglycerides. SI conversion factors: To convert TG to millimoles per liter, multiply by.113; cholesterol to millimoles per liter, by.259; uric acid to micromoles per liter, by 59.48. *Data are mean ± SD unless otherwise indicated. or more were 3% with fasting levels and 11% with nonfasting levels. Baseline characteristics for these men are shown in Table 1 by TG level for fasting and nonfasting conditions. The associations of each baseline characteristic with fasting TG level were similar to those of each characteristic with nonfasting TG level, except for age, alcohol use, and smoking. Age was negatively associated, and alcohol and smoking were positively associated, with natural log fasting TG level (P.1,.1, and.7, respectively), but none of the three was significantly associated with natural log nonfasting TG level (P=.42,.82, and.68, respectively). The Pearson correlation coefficient for fasting-nonfasting TG level was.67 on the original scale and.65 on the natural log scale. ASSOCIATIONS WITH CHD MORTALITY DURING 25 YEARS OF FOLLOW-UP Median follow-up from randomization through December 31, 1999, was 25.4 years, with 328 CHD deaths. Among the 874 men with fasting TG levels of 2 or more, there were 117 CHD deaths (13.4%); for those with fasting TG levels less than 2, there were 211 CHD deaths (1.9%). Among the 1724 men with nonfasting TG levels of 2 or more, there were 226 CHD deaths (13.1%); for the 185 men with nonfasting TG levels less than 2, 12 died of CHD (9.4%). Cumulative Probability of Death 2 15 1 5 Fasting TG <2 Fasting TG 2 5 1 15 2 25 3 Years Since Randomization Figure 1. Cumulative mortality from coronary heart disease among 289 Multiple Risk Factor Intervention Trial participants with both a baseline fasting and nonfasting triglyceride (TG) measure, by screen 2 fasting triglyceride level less than 2 vs greater than or equal to 2 (2.26 mmol/l). Plots of cumulative mortality for those with TG levels of 2 or more and less than 2 (Figure 1 and Figure 2) showed an increasing separation in risk from about year 7 for fasting and year 1 for nonfasting. Results from univariate- and multivariate-adjusted proportional hazards regression models are shown in Table 2. Both fasting and nonfasting TG levels were significant predictors of CHD mortality in unadjusted and 179

Cumulative Probability of Death 2 15 1 5 Nonfasting TG <2 Nonfasting TG 2 5 1 15 2 25 3 Years Since Randomization Figure 2. Cumulative mortality from coronary heart disease among 289 Multiple Risk Factor Intervention Trial participants with both a baseline fasting and nonfasting triglyceride (TG) measure, by screen 1 nonfasting triglyceride level less than 2 vs greater than or equal to 2 (2.26 mmol/l). in multivariate-adjusted (excluding HDL-C level) analyses, where hazard ratios (HRs) were similar for fasting and nonfasting TG levels. The strength of the relationships decreased with adjustment for HDL-C level, but HRs for 2 or more compared with less than 2 were still similar for fasting (HR, 1.24; P=.9) and nonfasting (HR, 1.26; P=.7) TG levels. In models with TG levels entered on a natural log (ln) scale, both fasting and nonfasting TG levels were again equally strong unadjusted and multivariate-adjusted (excluding HDL-C level) predictors of CHD mortality, and were weaker when further adjusted for HDL-C level. Multivariate-adjusted (excluding HDL-C level) HRs for CHD mortality for a 1-ln higher TG level (eg, 25 vs 92 mg/ dl) were 1.43 (P=.1) for fasting and 1.37 (P=.2) for nonfasting TG level. After adjustment for all confounding variables including HDL-C level, HRs dropped to 1.24 (P=.9) for fasting and 1.23 (P=.6) for nonfasting TG level. Education, income, employment status, and marital status were not significant predictors at level.5 in any of these models and were dropped. The interactions of TG level with HDL-C level, LDL-C level, systolic blood pressure, and number of cigarettes smoked per day were also considered and were not statistically significant. ASSOCIATIONS WITH NONFATAL OR FATAL CHD DURING 8 YEARS OF FOLLOW-UP There were 175 combined nonfatal or fatal CHD events, with median follow-up through February 28, 1982, of 7.6 years. Among the 874 men with fasting TG levels of 2 or more, there were 73 nonfatal or fatal CHD events (8.4%); for the 1935 with fasting TG level less than 2, 5.3% had a nonfatal or fatal CHD event. Among the 1724 men with nonfasting TG levels of 2 or more, there were 125 nonfatal or fatal CHD events (7.3%); for the 185 men with nonfasting TG level less than 2, 5 (4.6%) experienced a nonfatal or fatal CHD event. Plots of cumulative mortality for those with fasting and nonfasting TG levels of 2 or more and less than 2 showed an increasing separation in risk from about year 1 (not shown). Results from univariate and multivariate analyses are summarized in Table 3. Like the results for CHD mortality, both fasting and nonfasting TG levels were significant predictors of nonfatal or fatal CHD in unadjusted and multivariate-adjusted (excluding HDL-C level) analyses. Associations of TG level of 2 or more with nonfatal or fatal CHD were also significant after adjustment for HDL-C level: the HR was 1.64 (P=.4) for fasting and 1.46 (P=.3) for nonfasting. The multivariateadjusted HR for a 1-ln higher TG level was 1.56 (P=.8) for fasting TG level and 1.29 (P=.9) for nonfasting TG level. In general, HRs were larger for fatal or nonfatal CHD in 8 years than for fatal CHD in 25 years for both fasting and nonfasting TG levels. Education, income, employment status, and marital status were not significant predictors at level.5 in any of these models and were dropped. To assess whether the apparently stronger relationship of TG with nonfatal or fatal CHD in 8 years compared with CHD death in 25 years was due to the more proximal measurement of TG to the event, the 25-year follow-up period was divided into 3 periods with approximately equal numbers of CHD deaths: less than 12, 12 to 17, and 18 or more years after randomization, with 98, 13, and 127 deaths, respectively. Multivariateadjusted HRs for CHD mortality in the first 12 years associated with a TG level of 2 or more were 1.45 (95% confidence interval [CI],.92-2.28; P=.11) for fasting TG level and 1.8 (95% CI,.69-1.69; P=.73) for nonfasting TG level. In contrast, HRs for deaths after 18 years were.93 (95% CI,.61-1.41) for fasting TG level and 1.6 (95% CI, 1.5-2.45) for nonfasting TG level, indicating that nonfasting TG level may be more predictive of long-term mortality than fasting TG level. The HRs for deaths during 12 to 17 years were 1.51 (95% CI,.98-2.33) for fasting TG level and 1.14 (95% CI,.74-1.77) for nonfasting TG level. P values corresponding to the interaction of natural log transformed TG level and follow-up time were.99 for fasting TG level and.78 for nonfasting TG level. Taking these results together, then, there is no strong evidence that TG levels are more predictive of early (more proximal) CHD deaths than later deaths. BIVARIATE ASSOCIATIONS WITH FATAL CHD Analyses were also carried out classifying participants by both fasting and nonfasting TG levels simultaneously (Table 4). Most of the men with nonfasting TG level of 5 or more also had a fasting TG level of 2 or more (81%), and these men were at significantly increased risk of CHD death compared with men with both fasting and nonfasting TG levels less than 2 (HR, 1.59; P=.3). There were only 62 men with nonfasting TG level of 5 or more and fasting TG level less than 2 (19% of those with nonfasting TG level of 5 or more and 2% overall), among whom 5 CHD deaths occurred. Half of the men (144 [5%]) had nonfasting TG levels of 2 to 499 (2.26-5.63 mmol/l). For this subgroup, HRs were similar for those with high or with low fasting TG level: with fasting TG level of 2 or more, the HR (relative to men with both fasting and nonfasting TG level 18

Table 2. Hazard Ratios for 25-Year Coronary Heart Disease Mortality Among the 289 MRFIT Participants With Both a Baseline Fasting and Nonfasting Triglyceride Measure, Separately for Fasting and Nonfasting Triglycerides Unadjusted Multivariate Adjusted (No HDL-C)* Multivariate Adjusted (With HDL-C) HR 95% CI P Value HR 95% CI P Value HR 95% CI P Value Fasting triglycerides 2 vs 2 1.28 1.2-1.61.3 1.41 1.12-1.79.4 1.24.97-1.6.9 1-ln Higher TG 1.29 1.7-1.57.9 1.43 1.15-1.78.1 1.24.97-1.58.9 Nonfasting triglycerides 2 vs 2 1.49 1.17-1.89.1 1.4 1.1-1.78.6 1.26.98-1.62.7 1-ln Higher TG 1.32 1.1-1.59.3 1.37 1.12-1.67.2 1.23 1.-1.52.6 Abbreviations: BMI, body mass index; CI, confidence interval; HDL-C, high-density lipoprotein cholesterol; HR, hazard ratio; LDL-C, low-density lipoprotein cholesterol; ln, natural log; MRFIT, Multiple Risk Factor Intervention Trial; SBP, systolic blood pressure; TG, triglycerides. *Hazard ratios are adjusted for treatment group, age, LDL-C level, glucose level, SBP, cigarettes smoked per day, alcohol use, BMI, and African American vs non African American. Hazard ratios are adjusted for the foregoing plus HDL-C level. For example, 25 vs 92. Table 3. Hazard Ratios for 8-Year Nonfatal Plus Fatal Coronary Heart Disease Among the 289 MRFIT Participants With Both a Baseline Fasting and Nonfasting Triglyceride Measure, Separately for Fasting and Nonfasting Triglycerides Unadjusted Multivariate Adjusted (No HDL-C)* Multivariate Adjusted (With HDL-C) HR 95% CI P Value HR 95% CI P Value HR 95% CI P Value Fasting triglycerides 2 vs 2 1.64 1.21-2.22.1 1.84 1.35-2.51.1 1.64 1.17-2.29.4 1-ln Higher TG 1.5 1.16-1.95.2 1.56 1.12-2.16.8 1.56 1.12-2.16.8 Nonfasting triglycerides 2 vs 2 1.66 1.19-2.31.3 1.63 1.16-2.28.5 1.46 1.3-2.7.3 1-ln Higher TG 1.37 1.6-1.75.1 1.45 1.11-1.91.7 1.29.97-1.72.9 Abbreviations: BMI, body mass index; CI, confidence interval; HDL-C, high-density lipoprotein cholesterol; HR, hazard ratio; LDL-C, low-density lipoprotein cholesterol; In, natural log; MRFIT, Multiple Risk Factor Intervention Trial; SBP, systolic blood pressure; TG, triglycerides. *Hazard ratios are adjusted for treatment group, age, LDL-C level, glucose level, SBP, cigarettes smoked per day, alcohol use, BMI, and African American vs non African American. Hazard ratios are adjusted for the foregoing plus HDL-C level. For example, 25 vs 92. Table 4. Hazard Ratios for 25-Year Coronary Heart Disease Mortality Among the 289 MRFIT Participants With Both a Baseline Fasting and Nonfasting Triglyceride Measure, by Levels of Fasting and Nonfasting Triglycerides Nonfasting TG, Fasting TG, No. of Men No. (%) of CHD Deaths HR* 95% CI P Value 5 2 258 37 (14) 1.59 1.4-2.42.3 2 62 5 (8).82.33-2.6.68 2-499 2 517 69 (13) 1.35.96-1.9.8 2 887 115 (13) 1.25.94-1.67.12 2 2 99 11 (11) 1.31.69-2.46.41 2 986 91 (9) 1. Abbreviations: BMI, body mass index; CHD, coronary heart disease; CI, confidence interval; HDL-C, high-density lipoprotein cholesterol; HR, hazard ratio; LDL-C, low-density lipoprotein cholesterol; MRFIT, Multiple Risk Factor Intervention Trial; SBP, systolic blood pressure; TG, triglycerides. *HR is for each group relative to those with both fasting and nonfasting TG less than 2 ; HRs are adjusted for treatment group, age, HDL-C level, LDL-C level, glucose level, SBP, cigarettes smoked per day, alcohol use, BMI, and African American vs non African American. 2 ) was 1.35 (P=.8), and for men with fasting TG level less than 2, the HR was 1.25 (P=.12). This group of men with nonfasting TG level between 2 and 499 was at higher risk of CHD death regardless of their fasting TG level. ADJUSTMENT FOR REGRESSION DILUTION These associations of TG levels with CHD mortality and incidence are likely underestimated because of regression dilution. We first estimated the regression to the mean 181

Table 5. Estimation of Regression Dilution Based on All MRFIT Participants With Complete Annual Visit 2 Fasting Triglyceride and Baseline Risk Factor Data (Nonparametric Method) Screen 2 Fasting TG Quintile Mean Screen 2 Fasting TG, Adjacent-Quintile Mean Differences, Mean Second Annual Visit Fasting TG, Adjacent-Quintile Mean Differences, Ratio of Differences = Inflation Factor 1 (Lowest) 8.96... 15.52...... 2 121.6 4.64 137.34 31.82 1.28 3 159. 37.4 162.46 25.12 1.49 4 211.5 52.5 2.33 37.87 1.39 5 (Highest) 43.3 191.53 322.92 122.59 1.56 Abbreviations: MRFIT, Multiple Risk Factor Intervention Trial; TG, triglycerides. effect in the larger MRFIT cohort (N=11634) nonparametrically by computing mean screen 2 fasting TG levels and mean annual visit 2 fasting TG level within screen 2 quintiles. Ratios (screen 2 to annual visit 2) of quintilequintile differences in means ranged from 1.28 to 1.56 (Table 5). This indicated that an estimated regression coefficient for TG should be multiplied by up to 1.56 (inflated by up to 56%) to estimate the magnitude of the true association. The parametric estimate (inverse of the correlation coefficient between screen 2 fasting TG level and annual visit 2 fasting TG) was 1./.64=1.56, indicating again that TG regression coefficients should be inflated by up to 56%. This regression to the mean effect in TGs consequently affects the estimated levels of risk for CHD incidence and death, a phenomenon known as regression dilution bias. 2 The problem is possibly compounded by the selection (from the men screened for MRFIT) of highrisk men according to high cholesterol level for randomization into the MRFIT cohort. The estimated multivariate regression coefficient for a 1-ln higher fasting TG level was.213 (corresponding to a HR of 1.24; Table 3), so that the regression dilution adjusted HR was exp(.213 1.56)=1.39. Assuming the same regression dilution effect for nonfasting TG levels gives us similar results from the nonfasting TG regression coefficient of.27: exp(.27 1.56)=1.38. COMMENT The estimated risk of fatal CHD during 25 years was similar for fasting TG level of 2 or more vs less than 2 when compared with the risk of nonfasting TG level of 2 or more vs less than 2. This was true as well as for the risk associated with a 1-ln higher fasting TG level compared with nonfasting TG level. Associations of TG level with nonfatal or fatal CHD in 8 years were also similar for fasting and nonfasting TG levels and on both the dichotomized and the natural log scales. These associations tended to be stronger than for fatal CHD in 25 years, particularly for fasting TG level. However, there were no consistent trends indicating that the association with TG level, fasting or nonfasting, varied during follow-up; thus, the differences observed between these two end points could result from sampling variability. The estimated attenuation in the associations (with CHD mortality in particular) was substantial because of regression dilution. This attenuation is likely greater for nonfasting than fasting TG level because the intraindividual variability in nonfasting TG measurements would be expected to be higher. On the basis of Adult Treatment Panel III classification, the prevalence of hypertriglyceridemia (TG level 2 ) was double when nonfasting TG levels were used compared with the use of fasting TG levels (fasting prevalence, 31%; nonfasting prevalence, 61%), yet both were associated with a similar increase in risk of CHD. These two observations indicate that nonfasting TG may be useful for risk stratification. This was further supported by bivariate analyses of fasting and nonfasting TG, which, although limited by the number of events and therefore reduced power, suggested that elevated nonfasting TG may be associated with increased risk of CHD even among men with fasting TG level less than 2 mg/ dl. Use of a nonfasting TG level may be particularly appropriate for clinical trials where collection of a fasting blood sample is infeasible. No prospective studies, to our knowledge, have compared in the same cohort the strength of the association of fasting and nonfasting TG level with CHD risk. In a small case-control study, Patsch et al 17 found a stronger association of nonfasting than fasting TG level with coronary artery disease based on angiography. Comparisons of results across studies is hazardous because of variation in event definition and differential control of confounding factors. However, in a meta-analysis 18 of 17 prospective studies, an 89- (1-mmol/L) higher fasting TG level was associated with a 14% increased risk of cardiovascular disease after adjustment for HDL-C level and other risk factors. By comparison, in recent prospective studies of nonfasting TG level, stronger associations with CHD were noted. Stampfer et al 19 reported a 4% higher risk of myocardial infarction associated with a 1- (1.13-mmol/L) higher nonfasting TG level. Iso et al 2 reported a 26% (P=.9) increased risk of CHD associated with an 89- (1-mmol/L) higher nonfasting TG level. Schaefer et al 21 examined fasting plasma levels of TGs and remnantlike particles TG and their associations with carotid artery stenosis by ultrasound in the Framingham Offspring Study. No multivariate associations were found in men for TGs or remnantlike particles TG, whereas strong multivariate associations were found in women (P.1 for TGs and remnantlike particles TG). 182

The possible mechanisms by which TGs could affect CHD have been reviewed. 22-24 Ginsberg 23 noted that much of the evidence about TG and cardiovascular disease comes from studies in which TG level was measured in participants who were fasting. He noted that postprandial TG level and chylomicron remnants are predictive of cardiovascular disease 25,26 and speculated that the role of TG in atherogenesis might be more readily demonstrated with the use of nonfasting blood levels. Roche and Gibney 27 and Yu and Cooper 28 detailed the metabolic pathways for processing postprandial lipoproteins. Yu and Cooper described the lipolyzation of chylomicron TGs, allowing the delivery of free fatty acids to peripheral tissues. They described the roles of liver and liver receptors, the removal of chylomicron remnants from the blood, and the potential detrimental effects of chylomicron remnants on arterial walls. A limitation of our study is that we did not determine time since last meal for the nonfasting TG measurement. Furthermore, some of the differences between fasting and nonfasting TG levels could be due to serum vs plasma differences, and/or to regression to the mean (men at screen 1 were selected for further screening on the basis of elevated total serum cholesterol level, which is positively correlated with TG levels). In our cohort of 289, these screen 1 serum nonfasting values were on average 25% higher than screen 2 plasma fasting values. Previous work in the entire MRFIT cohort (12866 men) showed on average 4% higher serum than plasma TG levels from the same fasting blood sample. 29 It is a reasonable inference that further differences were predominantly due to nonfasting vs fasting state, even though we had no data on time of last meal for nonfasting samples. In conclusion, this work is, to our knowledge, the first direct comparison from the same study cohort of the prognostic importance of fasting and nonfasting TG levels for fatal and nonfatal CHD. Fasting and nonfasting TG levels were equivalently predictive both of 25-year CHD mortality and of 8-year fatal or nonfatal CHD on an unadjusted or multivariate-adjusted basis; adjustment for regression dilution made the predictiveness stronger. Taken together with the higher prevalence of hypertriglyceridemia with the use of nonfasting TG level, our results indicate that nonfasting TG level may be useful for risk stratification. Although multiple measurements may be required for accurate classification, nonfasting readings may provide information about risk that is not apparent from fasting readings alone. Accepted for publication August 9, 22. The Multiple Risk Factor Intervention Trial was conducted under contract with the National Heart, Lung, and Blood Institute, Bethesda, Md. This work was supported by National Heart, Lung, and Blood Institute grants R1-HL- 43232 and R1-HL-6814. This study was presented in part at the Fifth International Conference on Preventive Cardiology, Osaka, Japan, May 31, 21. Corresponding author and reprints: Lynn E. Eberly, PhD, Division of Biostatistics, School of Public Health, University of Minnesota, 42 Delaware St SE, MMC 33, Minneapolis, MN 55455-378 (e-mail: lynn@biostat.umn.edu). REFERENCES 1. NIH Consensus Development Panel on Triglyceride, High-Density Lipoprotein, and Coronary Heart Disease. NIH consensus conference: triglyceride, highdensity lipoprotein, and coronary heart disease. JAMA. 1993;269:55-51. 2. Clarke R, Shipley M, Lewington S, et al. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol. 1999;15:341-353. 3. Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Executive Summary of the Third Report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III).JAMA. 21;285:2486-2497. 4. Avins AL, Neuhaus JM. Do triglycerides provide meaningful information about heart disease risk? Arch Intern Med. 2;16:1937-1944. 5. Multiple Risk Factor Intervention Trial Research Group. Statistical design considerations in the NHLI Multiple Risk Factor Intervention Trial (MRFIT). J Chronic Dis. 1977;3:261-275. 6. Forum: the Multiple Risk Factor Intervention Trial (MRFIT): the methods and impact of intervention over four years. Prev Med. 1981;1:387-553. 7. Multiple Risk Factor Intervention Trial Research Group. Multiple Risk Factor Intervention Trial: risk factor changes and mortality results. JAMA. 1982;248:1465-1477. 8. Lipid Research Clinics Program, US Department of Health, Education and Welfare. Manual of Laboratory Operations, Volume 1. Washington, DC: National Institutes of Health; 1974. Publication DHEW (NIH) 75-628. 9. International Classification of Diseases, Ninth Revision, North American Clinical Modification. Vol 1. Ann Arbor, Mich: Edwards Brothers Inc; 1981. 1. Multiple Risk Factor Intervention Trial Research Group. Mortality after 16 years for participants randomized to the Multiple Risk Factor Intervention Trial. Circulation. 1996;94:946-951. 11. Horm J. Assignment of Probabilistic Scores to National Death Index Record Matches. Hyattsville, Md: National Center for Health Statistics; 1996. 12. Wentworth D, Neaton JD, Rasmussen W. An evaluation of the Social Security Administration Master Beneficiary Record File and the National Death Index in the ascertainment of vital status. Am J Public Health. 1983;73:127-1274. 13. World Health Organization. International Classification of Diseases, Tenth Revision. Geneva, Switzerland: World Health Organization; 1992. 14. Multiple Risk Factor Intervention Trial Research Group. Coronary heart disease death, nonfatal acute myocardial infarction and other clinical outcomes in the Multiple Risk Factor Intervention Trial. Am J Cardiol. 1986;58:1-13. 15. Cox DR. Regression models and life tables with discussion. J R Stat Soc Ser B. 1972;34:187-22. 16. Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika. 1994;81:515-526. 17. Patsch JR, Miesenböck G, Hopferwieser T, et al. Relation of triglyceride metabolism and coronary artery disease: studies in the postprandial state. Arterioscler Thromb. 1992;12:1336-1345. 18. Hokanson JE, Austin MA. Plasma triglyceride level is a risk factor for cardiovascular disease independent of high-density lipoprotein cholesterol level: a meta-analysis of population-based prospective studies. J Cardiovasc Risk. 1993;3:213-219. 19. Stampfer MJ, Krauss RM, Ma J, et al. A prospective study of triglyceride level, low-density lipoprotein particle diameter, and risk of myocardial infarction. JAMA. 1996;276:882-888. 2. Iso H, Naito Y, Sato S, et al. Serum triglycerides and risk of coronary heart disease among Japanese men and women. Am J Epidemiol. 21;153:49-499. 21. Schaefer EJ, Audelin MC, McNamara JR, D Agostino R, Wilson PW, and the Framingham Heart Study. Remnant-like particles and carotid atherosclerosis: results from the Framingham Offspring Study. Abstract presented at: American Heart Association Scientific Sessions 21; November 14, 21; Anaheim, Calif. Abstract 11556. 22. Grundy SM, Vega GL. Two different views of the relationship of hypertriglyceridemia to coronary heart disease. Arch Intern Med. 1992;152:28-34. 23. Ginsberg HN. Is hypertriglyceridemia a risk factor for atherosclerotic cardiovascular disease? a simple question with a complicated answer [editorial]. Ann Intern Med. 1997;126:912-914. 24. Harjai KJ. Potential new cardiovascular risk factors: left ventricular hypertrophy, homocysteine, lipoprotein(a), triglycerides, oxidative stress, and fibrinogen. Ann Intern Med. 1999;131:376-386. 25. Ginsberg HN, Jones J, Blaner WS, et al. Association of postprandial triglyceride and retinyl palmitate responses with newly diagnosed exercise-induced myocardial ischemia in middle-aged men and women. Arterioscler Thromb Vasc Biol. 1995;15:1829-1838. 26. Sharrett AR, Chambless LE, Heiss G, Paton CC, Patsch W. Association of postprandial triglyceride and retinyl palmitate responses with asymptomatic carotid artery atherosclerosis in middle-aged men and women: the Atherosclerosis Risk in Communities (ARIC) Study. Arterioscler Thromb Vasc Biol. 1995;15:2122-2129. 27. Roche HM, Gibney MJ. Effect of long-chain n-3 polyunsaturated fatty acids on fasting and postprandial triacylglycerol metabolism. Am J Clin Nutr. 2;71 (suppl):232s-237s. 28. Yu KC, Cooper AD. Postprandial lipoproteins and atherosclerosis. Front Biosci. 21;6:D332-D354. 29. Widdowson GM, Kuehneman M, DuChene AG, Hulley SB, Cooper GR. Quality control of biochemical data in the Multiple Risk Factor Intervention Trial: central laboratory. Control Clin Trials. 1986;7(suppl):17S-33S. 183