Non-HDL Cholesterol and Apolipoprotein B Predict Cardiovascular Disease Events Among Men With Type 2 Diabetes

The Health Professionals’ Follow-up Study is a prospective cohort study designed to study etiologies of heart disease, cancer, and other major chronic diseases in 51,529 U.S. male health professionals (dentists, veterinarians, pharmacists, optometrists, osteopathic physicians, and podiatrists) aged 40–75 years at study baseline in 1986 (8). Lifestyle factors and health outcomes have been obtained by questionnaires every 2 years. Dietary information using a food frequency questionnaire has been collected every 4 years. Between 1993 and 1994, 18,159 study participants provided blood samples. Among those with blood samples, 1,000 were diagnosed with definite type 2 diabetes at baseline or during follow-up until 1998 (with the vast majority being diagnosed before or within 2 years after blood collection). The present study included 746 diabetic men who did not report a diagnosis of angina pectoris, myocardial infarction, coronary bypass surgery, or stroke on any of the biennial questionnaires before blood collection.

Based on the diagnostic criteria proposed by the National Diabetes Data Group (9), a diagnosis of diabetes was established when at least one of following criteria was reported on a supplementary questionnaire sent to all men who reported a diagnosis of diabetes on any biennial follow-up questionnaire: 1) one or more classic symptoms (excessive thirst, polyuria, weight loss, hunger, or coma) plus a fasting plasma glucose concentration of ≥140 mg/dl or a random plasma glucose concentration of ≥200 mg/dl; 2) at least two elevated plasma glucose concentrations on different occasions (fasting ≥140 mg/dl and/or random ≥200 mg/dl and/or ≥200 mg/dl after 2 h or more on oral glucose tolerance testing) in the absence of symptoms; or 3) treatment with hypoglycemic medication (insulin or oral hypoglycemic agents). We used the National Diabetes Data Group criteria to define diabetes because our cases were diagnosed before the release of the American Diabetes Association criteria (10). Men with type 1 diabetes were excluded. A validation study in a subsample of the Health Professionals’ Follow-up Study demonstrated that our supplementary questionnaire is highly reliable in confirming diabetes diagnosis (11). Among a random sample of 71 men classified by our criteria as having type 2 diabetes according to the information reported on the supplementary questionnaire, medical records were available for 59. A physician blinded to the information reported on the questionnaire reviewed the records. The diagnosis of type 2 diabetes was confirmed in 57 (97%) of the 59 men.

The primary outcome of CVD consisted of fatal coronary heart disease (CHD), nonfatal myocardial infarction, fatal stroke, nonfatal stroke, coronary artery bypass, or angioplasty. We asked all men who reported a new diagnosis of nonfatal myocardial infarction or stroke on any biennial follow-up questionnaire to confirm the report and to provide permission to review medical records. Study physicians blinded to questionnaire information for the study participants reviewed the records. Myocardial infarction was confirmed using World Health Organization criteria, requiring symptoms plus either diagnostic electrocardiographic changes or elevated cardiac enzymes (12). Stroke was confirmed using the criteria of the National Survey of Stroke, which require a constellation of neurological deficits, sudden or rapid in onset, lasting at least 24 h or until death (13). Deaths were reported by next of kin, coworkers, or postal authorities or ascertained by a search for nonrespondents using the National Death Index; following up for mortality using these procedures exceeded 98% in a similar cohort study (14).

We sent a phlebotomy kit (including blood tubes, needles, a tourniquet, etc.) and instructions to men willing to provide blood specimens. The participants made arrangements for the blood to be drawn. Blood specimens were collected in three 10-ml liquid EDTA blood tubes, placed on ice packs stored in Styrofoam containers, and returned by overnight mail in a frozen water bottle. After arrival, the blood samples were centrifuged and stored in liquid nitrogen until laboratory analysis. Ninety-seven percent of samples arrived within 26 h of phlebotomy. We requested information on the date and time of the blood sample drawing and the time elapsed since the last meal to identify fasting subjects (>8 h). Quality control samples were routinely frozen along with study samples to monitor long-term stability of plasma samples collected and stored under this protocol. Frozen plasma aliquots from study subjects were selected for simultaneous analysis in 2002 and were analyzed in random order to reduce systematic bias and interassay variation.

The concentrations of total cholesterol, HDL cholesterol, and triglycerides were simultaneously measured on the Hitachi 911 analyzer using reagents and calibrates from Roche Diagnostics (Indianapolis, IN); coefficients of variation for these measurements were <1.8%. LDL cholesterol concentrations were measured by a homogenous direct method from Genzyme (Cambridge, MA). (The LDL cholesterol concentrations calculated using the Friedewald formula were similar to those of the actual measured LDL cholesterol concentrations.) The day-to-day variability at LDL cholesterol concentrations of 90, 106, and 129 mg/dl was <3.1%. ApoB concentrations were measured via an immunonephelometric assay using reagents and calibrators from Wako (Wako Chemicals, Richmond, VA) with a day-to-day variability of <5%. HbA_1c concentrations were determined based on turbidimetric immunoinhibition using hemolyzed whole blood or packed red cells. The day-to-day variability at HbA_1c concentrations of 5.5 and 9.1% was 1.9 and 3.0%, respectively. We calculated non-HDL cholesterol as the difference between total and HDL cholesterol.

Participants provided information biennially on their body weight, cigarette smoking, aspirin use, and physical activity. If the weight was missing we used the weight reported on the preceding questionnaire instead. Self-reports of body weight have been shown to be highly correlated with technician-measured weights (r = 0.96) in a parallel cohort of female health professionals (15).

We calculated BMI as the ratio of weight (in kilograms) to height (in meters squared), the latter being assessed in 1986 only. Physical activity (MET hours per week) was calculated using the reported time spent on various activities, weighting each activity by intensity level. History of high blood pressure was determined from self-reports preceding the blood collection. Family history of CHD was reported in 1986. Alcohol intake was estimated with a dietary questionnaire in 1994. Participants also provided information biennially on whether they were regularly using cholesterol-lowering drugs.

Each participant contributed follow-up time from the reported date of blood draw to the date of first event of CVD or death or 1 December 2000. We first calculated means ± SDs, medians, and proportions of potential CVD risk factors at baseline for 746 diabetic men who did and did not develop CVD during follow-up. Student’s t tests and χ² tests were used for comparisons of the means and the proportions. We divided the distributions of the lipid parameters into quartiles, and quartile-specific relative risks (RRs) of CVD were estimated from Cox proportional hazards models stratified on the 5-year age categories. In multivariate models, we adjusted for conventional CVD risk factors including BMI, family history of myocardial infarction, physical activity, cigarette smoking, alcohol consumption, fasting status (≥8 h), hypertension at baseline, aspirin use, and HbA_1c. All of these variables were included in the models as binary indicator variables. To reduce confounding due to lipid-lowering treatment, we further restricted the analyses to those who were not receiving cholesterol-lowering medications. Tests for trend were conducted by modeling the median of categories of each exposure as a continuous variable. Likelihood ratio tests were used for comparisons of nested models. We also calculated the area under the receiver-operating characteristic (ROC) curve to compare the models’ ability at discriminating between subjects who do and those who do not experience CVD events (16). The greater the area under the ROC curve, the better the predictivity of the model. Generally, an area under the ROC of 0.5 suggests no discrimination, whereas a maximal ROC of 1 suggests outstanding discrimination. All P values were two sided. All analyses were performed with SAS version 8.12 software (SAS Institute, Cary, NC).

In this cohort of 746 diabetic men, 19.2% reported using insulin. In addition, 39.3% reported taking sulfonylureas and 16.1% taking other diabetes medications (metformin, thiazolidinediones, and other medications). The distributions of potential CVD risk factors at baseline among the diabetic men who did and did not develop CVD during follow-up are presented in Table 1. Overall, diabetic men who developed CVD (18 myocardial infarction, 15 stroke, and 70 coronary procedures) were older and tended to have a history of hypertension compared with those who did not develop CVD. Compared with those who did not develop CVD, diabetic men with incident CVD had higher concentrations of non-HDL cholesterol, apoB, LDL cholesterol, total cholesterol, triglycerides, and higher ratio of total to HDL cholesterol and lower concentrations of HDL cholesterol. Distributions of other covariates were not statistically significant different between the two groups.

During the 6 years of follow-up, 103 CVD events occurred in 746 diabetic men. All lipid parameters were strongly positively associated with CVD risk in men with type 2 diabetes, except for HDL cholesterol, which was inversely associated with CVD risk (Table 2). In age-adjusted Cox proportional hazard models, the RRs (the highest quartile versus the lowest quartile) for non-HDL cholesterol (RR 2.25, 95% CI 1.24–4.08) and apoB (2.31, 1.25–4.27) were higher than those for LDL cholesterol, low HDL cholesterol, total cholesterol, and triglycerides but were lower than those for the ratio of total to HDL cholesterol and fasting triglycerides. After adjusting for BMI, family history of myocardial infarction, physical activity, cigarette smoking, alcohol consumption, fasting status, history of hypertension, aspirin use, and HbA_1c, the associations of these lipid parameters with CVD risk did not appreciably change (Table 2). Further analyses restricted to diabetic men not receiving cholesterol-lowering medications yielded similar results. Among those who did not receive cholesterol-lowing medications, the multivariate RR comparing the highest versus lowest quartile was 2.23 (95% CI 1.13–4.40) for non-HDL cholesterol, 2.34 (1.19–4.61) for apoB, 1.64 (0.87–3.09) for LDL cholesterol, 0.53 (0.28–0.99) for HDL cholesterol, 1.85 (0.95–3.59) for total cholesterol, 1.97 (1.03–3.72) for triglycerides, 3.32 (1.23–8.98) for fasting triglycerides, and 4.09 (1.96–8.51) for the total cholesterol–to–HDL cholesterol ratio.

To determine whether the predictive value of non-HDL cholesterol and apoB was independent of LDL cholesterol, we conducted log likelihood ratio tests by comparing two nested models, one model with LDL and other CVD risk factors and the other model with LDL and other CVD risk factors plus non-HDL cholesterol or apoB. The test results indicate that non-HDL cholesterol, but not apoB, adds significantly to the prediction of CVD risk beyond LDL cholesterol and other CVD risk factors in the multivariate model (χ² for adding non-HDL cholesterol = 6.90, P = 0.01; χ² for adding apoB = 0.80, P = 0.37).

We further compared the predictive value of non-HDL cholesterol or apoB among diabetic individuals with triglycerides levels <200 and ≥200 mg/dl. The associations of non-HDL cholesterol with CVD risk or apoB with CVD risk were stronger among diabetic men with lower triglyceride levels than those with higher triglyceride levels. However, tests for interaction were not statistically significant (P > 0.05) (data not shown).

We also plotted the ROC curves for LDL cholesterol, apoB, non-HDL cholesterol, and the total/HDL cholesterol deciles (Fig. 1). The area under the ROC curve for the CVD risk-prediction model with non-HDL cholesterol or apoB was larger than that for LDL cholesterol, indicating a better risk prediction for non-HDL cholesterol and apoB than for LDL cholesterol. The area under the ROC curve was largest for the total cholesterol–to–HDL cholesterol ratio, indicating the model with this ratio was superior to the other CVD risk-prediction models. The area with the ROC curve for the total cholesterol–to–HDL cholesterol ratio was also larger than that for the ratio of LDL to HDL cholesterol (0.697).

In this prospective cohort of 746 diabetic men aged 46–81 years with no previous history of CVD at the beginning of this study (time of blood draw), higher levels of non-HDL cholesterol and apoB at baseline were associated with a significantly increased future risk of CVD independent of BMI and other lifestyle CVD risk factors. Non-HDL cholesterol and apoB appeared to be stronger predictors of CVD risk than LDL cholesterol in patients with type 2 diabetes. The joint effect of total cholesterol and HDL cholesterol (the total cholesterol–to–HDL cholesterol ratio) was most strongly associated with CVD risk.

Non-HDL cholesterol has also been shown to be a strong predictor of CVD risk in patients with type 2 diabetes in two other studies. In the Strong Heart Study cohort (17), non-HDL cholesterol was a strong predictor of CVD in 2,108 American Indian men and women aged 45–74 years with diabetes during an average of 9 years of follow-up. The hazard ratios for the highest tertile of non-HDL cholesterol in diabetic men and women (2.23 and 1.80, respectively) were higher than those for either LDL cholesterol or triglycerides alone, although the CIs were overlapping. In a Finnish cohort study (18) of 1,059 middle-aged men and women with type 2 diabetes, higher levels of non-HDL cholesterol, as well as low HDL cholesterol and triglyceride levels, were each independently associated with a twofold increase in the risk of CHD morbidity or mortality during 7 years of follow-up. The association between LDL and CHD morbidity was weaker, and LDL cholesterol was not a significant predictor of CHD mortality. However, in the study, no direct comparison was made between LDL cholesterol and non-HDL cholesterol for the predictive values of CHD. Our findings that non-HDL cholesterol significantly predicts CVD risk beyond LDL cholesterol in diabetic men provide strong evidence supporting that non-HDL cholesterol could be a useful marker in CVD risk assessment in diabetic patients. Although several studies indicate that elevated apoB predicts an RR for CHD more than LDL cholesterol in the general population (19), to our knowledge there have been no previous population studies evaluating the value of apoB in predicting CVD risk in diabetic individuals.

Non-HDL cholesterol or apoB may be superior to LDL cholesterol in diabetic patients for several reasons (2,3). First, diabetes is often concomitant with many lipid and lipoprotein abnormalities characterized mainly by elevated VLDL, IDL, and chylomicrons, all of which are potentially atherogenic. The use of only LDL cholesterol ignores the significant contribution of atherogenic VLDL and IDL cholesterol to CVD. Second, the LDL cholesterol level is usually calculated from the Friedewald formula based on the measurement of total cholesterol, HDL cholesterol, and triglycerides (though in our study it was measured directly) (20). However, the Friedewald formula requires a fasting triglyceride level <400 mg/dl in order to accurately calculate LDL cholesterol. Because of elevated triglyceride levels in diabetic patients, the calculated LDL cholesterol level is likely to be unreliable. On the other hand, non-HDL cholesterol can be easily calculated from the total and HDL cholesterol levels without the limitation of hypertriglyceridemia, thus also eliminating the cost for additional lipid measurements. Furthermore, neither non-HDL cholesterol nor apoB measurements require fasting samples. In clinical practice, non-HDL cholesterol level can be used as a reasonable surrogate for apoB because of their high correlation and similar predictive role. Non-HDL cholesterol provides an even simpler and more familiar approach to quantify all atherogenic lipoproteins, although the methodology for measuring apoB is improving and becoming more widely available (2).

Consistent with previous studies of largely nondiabetic populations (21–25), the ratio of total to HDL cholesterol is the strongest lipid predictor for CVD in our diabetic cohort. This is not surprising because the ratio involves a joint or synergistic effect of two variables that have opposite biological effects. Of note, the total cholesterol–to–HDL cholesterol ratio is mathematically equivalent to the ratio of non-HDL to HDL cholesterol.

One limitation of our study was the relatively small sample size, particularly for fasting samples. For example, in the analyses comparing lipid parameters for predicting CVD risk, the RR comparing the highest versus the lowest quartiles of triglycerides was lower than the corresponding RR for non-HDL or apoB but was higher when limited to those who were fasting. The analyses of fasting samples should be interpreted with caution because the reference group (the lowest quartile) for triglycerides in the fasting sample had only seven CVD events, which could lead to unstable estimates. Also, we were unable to examine CHD and stroke separately but used a single outcome that combined the two. Nonetheless, the analyses with and without stroke as an end point yielded essentially the same results. Another issue deserving attention is that the diagnostic criteria for type 2 diabetes used in this study were changed in 1997 so that a lower fasting glucose level (≥126 mg/dl) would now be considered the diagnostic cutoff point. However, we believe that our results would apply to most diabetic patients because only a small fraction of diabetes cases would meet the newer American Diabetes Association criteria but not the National Diabetes Data Group criteria (fasting glucose between 126 and 140 mg/dl).

In conclusion, our data show that non-HDL cholesterol and apoB are more potent predictors of CVD incidence than LDL cholesterol among diabetic men. The potential value of using non-HDL cholesterol or apoB instead of LDL cholesterol as a primary target for cholesterol-lowing therapy in diabetic patients needs to be investigated further.

This work was supported by research grants (HL65582, HL35464, and CA55075) from the National Institutes of Health. F.B.H. is partly supported by an Established Investigator Award from the American Heart Association.