Joint Distribution of Non-HDL and LDL Cholesterol and Coronary Heart Disease Risk Prediction Among Individuals With and Without Diabetes

The data sets used in this study are all prospective cohort studies designed to study CVD and its risk factors. We confined our consideration to white men and women because of a limited number of blacks and other minorities, and we focused on participants aged ≥30 and free of CHD at baseline. The details of each of the following studies, in terms of study design and data collection, have been described elsewhere: FCS (8), FOS (9), LRCF (10), and MRFIT (11). A brief summary of each study is shown in online appendix 2 (available from http://care.diabetesjournals.org). Although each study used different questionnaires, similar questions were asked to assess each participant’s lifestyle factors and health outcomes. Diabetes status was defined as either “reported by physician-diagnosed and on medication” or having a fasting glucose level ≥126 mg/dl at the baseline examination based on the ADA’s new definition of diabetes (12).

The glucose and lipid profile were determined from fresh plasma after overnight fast (≥9 h). Glucose level was determined using either the techniques described by Cooper (13) or those of Bathelmai and Czok (14), but they were comparable. Total cholesterol was assayed by the Abell-Kendall method (15) in the FCS and FOS and by a Technicon AutoAnalyzer in the LRCF and MRFIT. The values of total cholesterol in the LRCF and MRFIT were standardized to the Abell-Kendall method. HDL cholesterol in all studies was assayed by manganese-heparin precipitation. Triglycerides were measured using the serum automated Lederer-Kessler method (16) in the FCS and FOS, but a calibration adjustment was applied to the measurements in the lipid study to match the average levels obtained by the Lipid Standardization Laboratory (17). In the LRCF and MRFIT, a preparation of an isopropanol extract of plasma was treated with a zerolite mixture to remove phospholipids, glucose, and bilirubin, and then plasma triglyceride levels were estimated fluorimetrically. All methods of triglyceride measurement are comparable because they were all standardized to the Lipid Research Clinics methods. LDL cholesterol in the FCS and FOS was estimated indirectly by use of the Friedewald formula when triglycerides were <400 mg/dl (6) and was estimated directly for triglycerides ≥400 mg/dl after ultracentrifugation of plasma and measurement of cholesterol in the bottom fraction (plasma density <1.006). In the LRCF, LDL cholesterol was determined by ultracentrifugation. The level of LDL cholesterol in the MRFIT was estimated indirectly using the Friedwald formula when triglycerides were <300 mg/dl and estimated directly by ultracentrifugation for triglycerides ≥300 mg/dl. Non-HDL cholesterol is defined as the difference between total and HDL cholesterol.

In the FCS and FOS, the end point diagnoses were based on a review of medical records by a committee of the FCS investigators. The diagnosis of CHD included fatal and nonfatal myocardial infarction, sudden cardiovascular death, and acute coronary insufficiency (8). In the LRCF, each participant’s vital status was followed prospectively to provide data on subsequent mortality. The protocol included annual mail and telephone contact with participants, but there was no clinical reexamination or assessment of morbidity. When a death was discovered, a copy of the death certificate was obtained, and the attending physician and next of kin were interviewed. Two members of a panel of five cardiologists, blinded to the participant’s identity and baseline characteristics, coded the cause of death as CHD, CVD (which included CHD), or “other.” Disagreements between the two cardiologists were settled by the entire panel (18). In the MRFIT, the vital status of each man was checked by clinical center staff during the trial and at the termination of active intervention in February 1982. Then, vital status was determined from the National Death Index and Social Security Administration. Cause-specific death rates are based on coding of death certificates by trained nosologists, using the ICD-9 (19). Two nosologists independently coded each death certificate, and a third nosologist adjudicated any disagreement (20). Using the ICD-9, deaths from CHD were denoted by codes 402, 410–414, and 429.2 (21).

Age at baseline was calculated by the date of birth and the date of exam when plasma lipoproteins were measured for each cohort. Smoking status was defined as current cigarette smokers and noncurrent smokers based on self-report from the baseline questionnaire. BMI was calculated as the ratio of weight (in kilograms) to height (in meters squared) measured at the baseline examination. Systolic blood pressure (in mmHg) was the average of two readings at the baseline examination.

Follow-up time was calculated as the difference between the date of end point and the date of blood draw from each cohort. The end point was defined as either death from CHD or end of study, which varies from cohort to cohort. Means and proportions were calculated for the CHD risk factors at baseline based on diabetes status. Student’s t tests and χ² tests were used for comparisons of means and the proportions. Because of the limited number of individuals with diabetes in each cohort, only pooled analyses were performed. Based on ATP III guidelines, individuals with diabetes are defined as a CHD risk equivalent, and the therapeutic goal for dyslipidemia management is set as an LDL cholesterol level <100 mg/dl. For non-HDL cholesterol, the cut points are 30 mg/dl higher than that for LDL cholesterol to account for the VLDL cholesterol fraction. Therefore, we divided the distributions of these two lipid parameters in the following way: LDL cholesterol: <100, 100–129, and ≥130 mg/dl; and non-HDL cholesterol: <130, 130–159, and ≥160 mg/dl. For other lipid parameters, we followed the recommendations of the ATP III guidelines as follows: total cholesterol: <200, 200–239, and ≥240 mg/dl; HDL cholesterol: ≥60, 40–59, and <40 mg/dl; triglycerides: <150, 150–199, and ≥200 mg/dl. The risk of CHD death within each level of the various lipid parameters was estimated for individuals with and without diabetes. The lowest category of each lipid parameter from those without diabetes, except for HDL cholesterol, where the highest from those without diabetes was used, was used as the reference group. As a result, five indicator variables were created for each individual in the data set. For example, the reference group for LDL cholesterol was individuals with LDL <100 (mg/dl) and without diabetes, and the five indicator variables were: 1) LDL 100–129 and without diabetes (1 = yes, 0 = no); 2) LDL ≥130 and without diabetes (1 = yes, 0 = no); 3) LDL <100 and with diabetes (1 = yes, 0 = no); 4) LDL 100–129 and with diabetes (1 = yes, 0 = no); and 5) LDL ≥130 with diabetes (1 = yes, 0 = no). For the joint distribution of non-HDL and LDL cholesterol, four groups were created based on lipid levels (in mg/dl): 1) non-HDL <130 and LDL <100; 2) non-HDL <130 and LDL ≥100; 3) non-HDL ≥130 and LDL <100; and 4) non-HDL ≥130 and LDL ≥100.

Group 1 within those without diabetes was the overall reference group, and seven indicator variables were created. Cox proportional hazards models were used in three approaches to estimate the relative risk (RR) of CHD death for non-HDL and LDL with adjustment for age, cohort study, current smoking status, BMI, systolic blood pressure, and sex (when applicable). First, both non-HDL and LDL were added into the model as continuous variables, and their impacts on CHD death were estimated by diabetes status and by sex. Second, non-HDL and LDL were retained as continuous variables, and diabetes status was included as well as interaction terms between diabetes status and non-HDL and between diabetes status and LDL. Finally, the RRs for grouped non-HDL, LDL, and other lipids were estimated. We also conducted receiver operating characteristic analyses and compared the c-statistics for the overall predictive value of non-HDL and LDL as a continuous variable separately as well as the category variables separately in the model, with adjustment for the covariates mentioned above (22). All analyses were performed using SAS statistical software, version 8.2 (SAS Institute, Cary, NC).

There were 12,660 men and 6,721 women from the four cohorts included in this analysis (Table 1). The proportion of individuals with diabetes is higher in the FCS than in the other cohorts. Overall, ∼6% of men and ∼4% of women were categorized as having diabetes status.

Overall, individuals with diabetes were older and were more likely to be male; to have higher levels of total, VLDL, and non-HDL cholesterol and triglycerides; to have lower levels of HDL cholesterol; to have higher levels of systolic blood pressure; and to have a larger BMI (Table 2). Except for current smoking status and LDL cholesterol levels, all differences between individuals with and without diabetes were statistically significant.

During the average ∼13 years of follow-up, 114 CHD deaths occurred in 1,018 individuals with diabetes, and there were 659 CHD deaths among the 18,363 individuals without diabetes. Non-HDL and LDL were initially considered in the Cox regression model as continuous variables and examined by diabetes status with adjustment for age, sex, study, BMI, systolic blood pressure, and current smoking status. Among those without diabetes, each increase by 1 mg/dl of non-HDL cholesterol was associated with a 5% increased risk for CHD death (95% CI 1.001–1.008, P < 0,001), and every 1-mg/dl increase of LDL cholesterol was associated with a 4% increased risk for CHD death (1.001–1.008, P = 0.02). These results were nearly identical among those with diabetes. When stratified by sex and diabetic status, both men and women had similar RRs for CHD from non-HDL and LDL. These latter findings reached statistical significance only for non-HDL among men (P = 0.005 for those with diabetes and P = 0.02 for those without diabetes) and for LDL among those without diabetes (P = 0.02). There was no evidence of an interaction between diabetic status and lipid level; the RR estimates were ∼1.0 for the interaction terms (P > 0.5). Overall, the c-statistics from ROC analysis for non-HDL and LDL was similar (0.799 vs. 0.796 as continuous variables; 0.798 vs. 0.797 using the cut points previously described).

When LDL, non-HDL, and other lipid measurements were considered as categorical variables as described above (see research design and methods), all lipid parameters were strongly positively associated with the risk for CHD death among individuals with and without diabetes, except for HDL cholesterol, which was negatively associated with the risk for CHD death (Table 3). The associations were stronger among individuals with diabetes; generally, the RRs for individuals with diabetes were ∼200% higher than that for individuals without diabetes in corresponding lipid levels.

The risk for CHD death for the joint distribution of non-HDL and LDL cholesterol by diabetes status is shown in Fig. 1 (see online Appendix 1 for the RR estimates and 95% CIs for each cell). Compared with the reference group (without diabetes with non-HDL <130 mg/dl and LDL <100 mg/dl), the RR (95% CI) for CHD death among those with diabetes did not increase with increasing level of LDL, whereas it did increase with increasing level of non-HDL: group 1: 5.7 (2.0–16.8); group 2: 5.7 (1.6–20.7); group 3: 7.2 (2.6–19.8); and group 4: 7.1 (3.7–13.6). Among those without diabetes, the risk for CHD death increased with higher levels for both non-HDL and LDL cholesterol: group 2: 3.3 (1.6–6.7); group 3: 2.3 (0.9–5.4); and group 4: 3.8 (2.0–7.1).

The results from this pooled prospective cohort confirmed that diabetes status is a strong risk factor for CHD death. Our results indicate that the categorical 130 mg/dl cutoff point for non-HDL appears to be a better predictor of the risk for CHD death than the 100 mg/dl cutoff point for LDL among those with diabetes, and it may serve as a useful clinical tool. The failure of the ROC analyses to find one lipid parameter superior to another may not be surprising because the interpretation of the RRs is, at least in part, dependent on the measurement scale, the cut points used, and the manner in which the variables are modeled (23).

Several other studies have shown that the level of non-HDL cholesterol was a stronger predictor of CHD or CVD risk among patients with diabetes. In the Health Professionals’ Follow-up Study (24), non-HDL cholesterol was a strong predictor of CVD in 746 diabetic men aged 46–81 years during a 6-year follow-up. The RR for the highest quartile of non-HDL cholesterol was significantly higher than that for LDL cholesterol: 2.34 (95% CI 1.26–4.43) vs. 1.74 (0.99–3.06). In a Finish cohort study (25), 1,059 middle-aged men and women with type 2 diabetes were followed for 7 years, and it was found that higher levels of non-HDL cholesterol were independently associated with a twofold increase in the risk for CHD death or morbidity, although there was no direct comparison between LDL and non-HDL cholesterol for the predictive values of CHD. In the Strong Heart Study cohort (26) of 2,108 American-Indian men and women aged 45–74 years with diabetes, the hazard ratios for the highest tertile of non-HDL cholesterol were higher than those for LDL cholesterol among both diabetic men and women, although the CIs were overlapping.

This is among the first studies to evaluate the predictive value of the joint distribution of non-HDL and LDL cholesterol among individuals with and without diabetes, although two of the three studies mentioned above performed some comparisons between LDL and non-HDL cholesterol. According to ATP III guidelines, for patients with diabetes, the goal for LDL cholesterol–lowering therapy is an LDL cholesterol level <100 mg/dl. For a baseline LDL cholesterol <100 mg/dl among those with diabetes, no further LDL cholesterol–lowering therapy was recommended. Evidence from the subgroup analysis of patients with diabetes from the Heart Protection Study (HPS) trial indicated that even among those diabetic patients with very low LDL cholesterol (<116 mg/dl) at entry, a marginally significant risk reduction of CHD event was observed with LDL cholesterol–lowering therapy (simvastain) (27). This suggested that the goal for LDL cholesterol therapy among patients with diabetes can be set even lower than currently recommended (28). Our results further suggest that non-HDL might be a better primary target for lipid reduction than LDL among patients with diabetes. The potential value of using non-HDL as the dyslipidemia management tool among patients with diabetes needs to be further investigated.

There are several reasons why the level of non-HDL may be superior to that of LDL in CHD risk prediction among individuals with diabetes (29). First, non-HDL cholesterol contains all potential atherogenic lipoproteins, including VLDL, intermediate-density lipoprotein, and LDL, whereas LDL cholesterol does not. The characterized dyslipidemia among diabetic individuals is elevated triglyceride-rich lipoproteins (VLDL and intermediate-density lipoprotein) and decreased HDL. The use of LDL alone will ignore the contribution of those triglyceride-rich lipoproteins in the development of CHD. Second, in the clinical lipoprotein analysis, the level of LDL cholesterol is usually estimated using Friedewald’s formula, based on the measurements of total and HDL cholesterol and triglycerides, which is used to estimate the value of VLDL cholesterol. However, the estimation of LDL by this formula becomes progressively less accurate as the triglyceride level increases, and the formula is no longer considered accurate enough for use when triglyceride levels reach 400 mg/dl. Because of elevated triglyceride levels in patients with diabetes, the level of LDL estimated by this formula is likely to be unreliable. In contrast, the level of non-HDL can be easily calculated from the difference between the levels of total and HDL cholesterol. In addition, there is no assumption about the composition of VLDL particles; thus, the non-HDL level can be calculated in the nonfasting state or in the setting of hypertriglyceridemia.

Among the main limitations of this study are that it is a post hoc analysis based on the new ADA definition of diabetes applied to this pooled cohort and also that only ∼11% of population met the ATP III goal for an LDL level <100 mg/dl, whereas 15% of the population experienced a non-HDL level <130 mg/dl. Thus, the results from this analysis may not be generalizable to the general population. However, the ATP III recommendations for screening and treatment were established for those with an elevated absolute risk of CHD, which the majority of participants in this analysis would most likely meet by today’s standards. There was no information for us to distinguish type 2 from type 1 diabetes, although the majority of these adults likely had type 2 diabetes, given its much higher prevalence in the general population (1). There are several significant strengths to this investigation, including the large sample size from this pooled cohort and also that each of the studies included was originally designed to examine the etiologies of CHD risk and that our end point (CHD death) was well documented.

In conclusion, these results suggest that non-HDL cholesterol level is a stronger predictor of CHD death than LDL cholesterol among those with diabetes, and they further suggest that VLDL cholesterol (and/or VLDL triglyceride) may play a critical role in the development of CHD among those with diabetes. These findings should be considered in the clinical approach to risk reduction among diabetic patients.

The Framingham Heart Study, LRCF, and MRFIT are conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with the study investigators.

This article was prepared using a limited-access dataset obtained from the NHLBI and does not necessarily reflect the opinions or views of the individual studies or the NHLBI.