Different Lipid Variables Predict Incident Coronary Artery Disease in Patients With Type 1 Diabetes With or Without Diabetic Nephropathy: The FinnDiane Study

Coronary artery disease (CAD) remains the leading cause of death in the general population (1). Patients with diabetes are particularly susceptible to atherosclerosis and premature CAD deaths (2,3). In type 1 diabetes, the additional CAD risk appears to be intrinsically connected to diabetic nephropathy, as patients with normal albumin excretion rate (AER) show similar mortality to the background population, whereas patients with macroalbuminuria or end-stage renal disease (ESRD) are 10 times more likely to die prematurely (4,5). Diabetic nephropathy has a substantial association with CAD incidence, but it may also simultaneously obscure other important risk factors (6,7). Therefore, it is important to examine the specific circumstances of type 1 diabetes and different stages of nephropathy in relation to established CAD risk factors in the general population.

Hypercholesterolemia is the classical causal risk factor for the development of atherosclerosis, (8) and lowering LDL cholesterol (LDL-C) continues to be the cornerstone of pharmacological interventions to reduce CAD mortality (9). On the other hand, uncomplicated type 1 diabetes is associated with a favorable lipid profile (10), whereas diabetic nephropathy is associated with a dyslipidemic pattern similar to that seen in patients with type 2 diabetes (6,11). It is thus uncertain if and how nephropathy modulates the lipid risk profile for CAD in type 1 diabetes and whether the lipid targets for CAD prevention in type 1 diabetes differ from those of the general population.

Therefore, we decided to assess the capability of the lipid variables, the apolipoproteins, and their ratios to predict incident CAD events in a large well-characterized cohort of type 1 diabetes. We will also explore how albuminuria, glycemic control, and sex modulate the risk profile.

The Finnish Diabetic Nephropathy Study (FinnDiane) has been previously prescribed (11,12). Estimated glomerular filtration rate (eGFR) was calculated on the basis of a serum creatinine measurement using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) formula (13). Nephropathy status was determined based on the measurement of AER in at least two out of three consecutive 24-h or overnight urine collections. AER <20 μg/min or <30 mg/24 h was defined as normal AER, 20≤ AER <200 μg/min or 30≤ AER <300 mg/24 h as microalbuminuria, and AER ≥200 μg/min or AER ≥300 mg/24 h as macroalbuminuria. ESRD was defined as requirement of dialysis or renal transplantation. In addition to the urine collections used for the classification of the renal status, 24-h AER was also measured centrally from a single sample with an immunoturbidimetric method. The result of this measurement was used in the multivariate analyses. Estimated glucose disposal rate (eGDR) was calculated with an equation developed by Williams et al. (14) modified for use with HbA_1c instead of HbA₁ (12). Serum lipid and lipoprotein concentrations were all measured centrally from serum samples using earlier described methods (11). LDL-C was calculated with the Friedewald formula if triglycerides were <4.0 mmol/L (15).

At baseline, patients underwent a thorough clinical investigation at a regular visit to their attending physician. The baseline data were collected between 1994 and 2008. Complete baseline data including centrally measured lipid profiles were available for 3,872 patients. Follow-up data were obtained from the Finnish Hospital Discharge Register (HDR) based on hospital discharge records and the Finnish Cause of Death Registry (CDR) through to 31 December 2010 and available for all patients. An incident CAD event was defined as myocardial infarction given as ICD-10 code I21 (ICD-9, 410), coronary artery bypass graft surgery or coronary angioplasty. Researcher N.T. from the FinnDiane study group verified the Finnish HDR data by double-checking the hospital records of 28% of patients. In this sample, no classification errors were found, and only four borderline cases of acute myocardial infarction were identified. Otherwise all cases were in accordance with the universal definition of myocardial infarction (16) or underwent either coronary artery bypass graft surgery or coronary angioplasty. Fatal CAD events were identified from a search of the Finnish CDR and established when the immediate or underlying cause of death was from CAD, i.e., given as ICD-10 codes I20–25 (ICD-9, 410–414). Death certificates were also obtained in order to verify the register data. Patients with acute myocardial infarction, coronary artery bypass graft surgery, or coronary angioplasty already at baseline were excluded from the study (n = 306). Furthermore, patients with ICD-10 diagnosis codes I20 and I22–25 (ICD-9, 411–414) in the Finnish HDR, with reported coronary heart disease, or taking long-acting nitroglycerin medication at baseline were excluded from the control group (n = 46). Hence, a total of 3,520 patients were included in the study.

Data for normally distributed and continuous variables are presented as mean ± SD and data for nonnormally distributed variables as median with IQR. Differences between groups were analyzed with Student t test, ANOVA, or Mann-Whitney U test as appropriate. Categorical variables were analyzed using Pearson χ² test. Nonnormally distributed values were logarithmically transformed before inclusion in the models. Univariate and multivariate Cox regression models were used to analyze the associations between CAD risk factors and incident CAD events. For the multivariate analyses, model selection from variables univariately associated with CAD was accomplished by minimization of the Akaike information criteria (AIC). Every variable reduced the AIC except for sex, but since it has previously been well established as a CAD risk factor, it was included in the models. The multivariate model used in the analyses included diabetes duration, eGFR, systolic blood pressure (SBP), retinal laser treatment, AER, sex, HbA_1c, waist-to-hip ratio (WHR), history of smoking, and one of the lipid variables. Because of collinearity, only one lipid variable at a time was entered into the models. Results are presented as hazard ratios (HRs) per one SD increase of the study cohort with 95% CIs. The standardized score for WHR was calculated separately for men and women.

To compare the Cox models, we calculated the area under the curve (AUC) of the receiver operating characteristics. Likelihood ratio (LR) χ² statistics from the Cox models were also calculated; a higher value indicates a better global fit. Net reclassification improvement (NRI) is the percentage reclassified after the inclusion of the lipid variable in the above-mentioned multivariate model, distinguishing movement in the correct direction, i.e., the proportion of subjects being reclassified to a higher risk category among CAD cases or a lower risk category among control subjects (17). In CAD prevention, the 5, 10, and 20% cutoff points have been proposed as relevant for clinical decision making (18,19) and were therefore chosen as NRI cutoff points. Fine and Gray (20) regression analysis, which extends the Cox proportional hazard model to competing risk data by consideration of the subdistribution hazard, was also performed to take into account the competing event of non-CAD death. After these analyses, figures of the cumulative incidence for CAD in normoalbuminuric or macroalbuminuric patients divided by the median of lipid variables were drawn. A more stringent P value of <0.01 was considered statistically significant in order to correct for multiple testing. Statistical analyses were performed using PASW Statistics 18 for Windows (SPSS Inc., Chicago, IL), STATA Data Analysis and Statistical Software (StataCorp LP, College Station, TX), or MedCalc (MedCalc Software BVBA, Ostend, Belgium).

During a follow-up period of 10.2 years (8.6–12.0), 310 (male/female, 173/137) out of 3,520 (9%) patients suffered an incident CAD event. The clinical characteristics of the patients stratified by an incident CAD event and non-CAD death can be seen in Table 1. In general, patients who had an incident CAD event were older and had longer diabetes duration, higher SBP, higher AER, lower eGFR, lower eGDR, and a higher frequency of antihypertensive, lipid-lowering, and retinal laser treatment than patients without a CAD event. The lipid profile in patients with an incident CAD event was characterized by higher total cholesterol, non–HDL cholesterol (non–HDL-C), LDL-C, triglycerides, and apolipoprotein B (ApoB) as well as lower HDL-C and HDL₂-C than in patients without CAD. Likewise, the lipid and apolipoprotein ratios were worse in patients with CAD than in patients without. The lipid profiles of patients stratified by their renal status can be seen in Supplementary Table 1.

In Cox regression analysis including the entire cohort, diabetes duration, eGFR, ApoB, SBP, and retinal laser treatment were independent predictors of an incident CAD event (Supplementary Table 1). The model also included sex, HbA_1c, WHR, AER, and history of smoking. Among the lipid variables, the highest HR per one SD increase of the study cohort was found for ApoB. When ApoB was replaced with all of the other lipid variables, one at a time, triglycerides, non–HDL-C, ApoB/ApoA-I, and triglyceride/HDL-C ratio were also good predictors of an incident CAD event (Table 2). In order to further compare the performance of different lipid variables in predicting an incident CAD event, we calculated the AUC of the different Cox regression models. The highest AUCs were found for ApoB/ApoA-I and triglyceride/HDL-C ratio followed by ApoB and triglycerides. In line, P values for NRI were <0.05 for ApoB, ApoB/ApoA-I ratio, and triglycerides, indicating that these parameters would be useful for risk stratification. Diabetes duration and age were highly correlated (r = 0.73) and were therefore not entered into the same model. In the univariate models, diabetes duration was more strongly associated with CAD, and it also decreased AIC more than age and was therefore chosen to be included in the multivariate models. If, however, age was added to the original Cox regression model, age was also a significant independent predictor (P = 0.0002) of CAD together with diabetes duration, eGFR, ApoB, and retinal laser treatment. When use of lipid-lowering agents was added to the original model, it was not a significant predictor of CAD (P = 0.10) and did not change the results. When patients with lipid-lowering medication were excluded from the analysis, diabetes duration, eGFR, and ApoB were still independent predictors of CAD (data not shown). When the competing event of non-CAD death was taken into account by performing Fine and Gray regression analyses, similar results were found as in the Cox regression analyses (Supplementary Table 1).

When men and women were analyzed separately, ApoB was an independent predictor of CAD in men (Supplementary Table 1). In women, triglyceride/HDL-C, ApoB/ApoA-I ratio, and triglycerides had the highest HR for an incident CAD event.

In patients with normal AER at baseline, ApoB/ApoA-I ratio, triglyceride/HDL-C ratio, triglycerides, and ApoA-I were all independent predictors of an incident CAD event (Table 3). In microalbuminuric patients, the number of events was only 41, and none of the lipid variables were significant predictors of CAD. In macroalbuminuric patients, ApoB had the highest HR of CAD followed by non–HDL-C, LDL-C, and total cholesterol. In patients with ESRD, non–HDL-C, total cholesterol, and ApoB/ApoA-I ratio reached borderline significance (P < 0.05).

In Cox regression models stratified by the median HbA_1c of the cohort (8.3%, 67 mmol/mol), ApoB had the highest HR for an incident CAD event in patients with an HbA_1c level above the median followed by total cholesterol, non–HDL-C, and LDL-C (Supplementary Table 1). However, these variables were poor predictors of CAD in patients with an HbA_1c level below the median. Instead ApoB/ApoA-I ratio, triglyceride/HDL-C ratio, and triglycerides were independent predictors of CAD in patients with HbA_1c below the median.

To look at the effect of clustering of risk factors, we divided patients into five groups according to the amount of risk factors present. In a Cox regression analysis, the HR was 3.27 in patients with three out of the five risk factors, 5.96 in patients with four risk factors, and 7.02 in those with all five risk factors compared with patients without any or with only one of the five risk factors (Supplementary Table 1).

Figure 1 and Supplementary Fig. 1 show the cumulative incidence of CAD in normoalbuminuric and macroalbuminuric patients divided by their median lipid levels. The median LDL-C and total cholesterol levels performed poorly in normoalbuminuric patients (Fig. 1A and Supplementary Fig. 1A), but in macroalbuminuric patients, they were better predictors together with non–HDL-C. When normoalbuminuric patients were divided by their median triglyceride level (0.94 mmol/L), triglycerides performed well (Fig. 1C), but when they were divided by the current guideline for triglycerides (<1.7 mmol/L) (9), they could not predict a future CAD event (Fig. 1E). In macroalbuminuric patients, the guideline cutoff level performed better (Fig. 1F). In normoalbuminuric patients, the median of triglyceride/HDL-C and ApoB/ApoA-I ratio were the best predictors for an incident CAD event (Supplementary Fig. 1I and K).

This study is the first report on a complete panel of both antiatherogenic and atherogenic lipid parameters as predictors of incident CAD outcomes in patients with type 1 diabetes. In this study we have shown that in patients with type 1 diabetes, several lipid and apolipoprotein variables are associated with an incident CAD event depending on the patient’s sex, glycemic control, or renal status. In normoalbuminuric patients, the ratios of concentrations of atherogenic and antiatherogenic lipoproteins and lipids were the strongest predictors of an incident CAD event, whereas in macroalbuminuric patients, no added benefit was gained from the ratios. The same phenomenon was seen when the patients were divided by sex or glycemic control.

Previous data are limited, but non–HDL-C, HDL-C, and triglycerides have been reported to be independent predictors of CAD in patients with type 1 diabetes (21,22). In this study, ApoB was the best independent predictor of an incident CAD event in the entire cohort followed by triglycerides, non–HDL-C, ApoB/ApoA-I ratio, and lipid ratios. Unexpectedly, LDL-C performed less well than the other lipid variables. The predictive performance of different variables for CAD outcomes was also evaluated by calculating NRI percentages where ApoB and ApoB/ApoA-I ratio performed best followed by triglycerides and total cholesterol/HDL-C ratio, outperforming LDL-C. ApoB is the main protein of the atherogenic particles VLDL, IDL, and LDL (23), and >90% of the circulating ApoB is located in LDL particles (24). Thus, ApoB is a measure of all ApoB-containing particles, including triglyceride-rich lipoproteins, their remnants, as well as LDL particles. It should be recognized that non–HDL-C is likewise capturing the risk associated with triglyceride-rich lipoproteins, their remnants, and LDL-C. Thus, the poorer performance of LDL-C when replaced by these variables is understandable. The antiatherogenic HDL-C and related parameters were also significant predictors of CAD outcomes but performed less well than atherogenic parameters, indicating an imbalance between antiatherogenic and atherogenic variables.

Diabetic nephropathy is strongly associated with dyslipidemia in patients with type 1 diabetes (6,11). Previous studies have found even more robust lipid disturbances associated with nephropathy than with cardiovascular disease (CVD) (25). To the best of our knowledge, other prospective studies have not taken into account the renal status when evaluating the prediction of lipid variables for an incident CAD event in patients with type 1 diabetes. When the patients were divided by their renal status, the ApoB/ApoA-I and triglyceride/HDL-C ratios had the highest HR of CAD in patients with normal AER at baseline. Total cholesterol and LDL-C were not independent predictors of CAD in normoalbuminuric patients. Figure 1A and Supplementary Fig. 1A illustrate how poorly they predicted an incident CAD event in multivariate models in normoalbuminuric patients. As expected, in macroalbuminuric patients, total cholesterol and LDL-C were elevated and they performed well as CAD predictors together with ApoB and non–HDL-C. Notably, HDL-C or related parameters were not significant predictors of CAD events in macroalbuminuric patients. In a case-control study of the EURODIAB cohort, CVD was not associated with any traditional lipid variables in normoalbuminuric patients. However, in macroalbuminuric patients, CVD was associated with increased triglyceride and LDL-C levels (25). In the EURODIAB study, the ApoB/ApoA-I or triglyceride/HDL-C ratios were not calculated and the CAD events were not analyzed separately.

HDL-C values are generally similar or even higher in patients with type 1 diabetes compared with the general population (10). The higher HDL-C concentrations can partly be explained by the subcutaneous insulin administration, which enhances lipoprotein lipase and is associated with an improved lipid profile (26,27). However, whether or not the functionality and protective effect of HDL-C are similar in patients with type 1 diabetes compared with that of the general population has been questioned (28). For example, the glycation of ApoA-I affects the reverse cholesterol transport ability of HDL particles (29), and the HDL of patients with type 1 diabetes protects erythrocyte membranes from oxidative damage less efficiently (30). Moreover, HDL particles failed to counteract the inhibitory effect of oxidized LDL on vasodilatation in patients with type 1 diabetes (31). In type 2 diabetes, increased concentrations of advanced glycation end products are associated with impairment in the antioxidant ability of HDL particles in patients with nephropathy (32), and the potency of HDL to inhibit cytokine release by macrophages is attenuated (33). In this study, ApoA-I and HDL-C were poor predictors of incident CAD events in the entire cohort, in patients with diabetic nephropathy and poor glycemic control and in men. This suggests that the protective effect of HDL and ApoA-I could be impaired, especially in patients with diabetic nephropathy or poor glycemic control.

LDL-C is the established target of therapy for the prevention of CAD (9,34) as lack of data from randomized clinical trials does not allow the definition of targets either for triglyceride-lowering or HDL-C–raising interventions. However, consensus exists that non–HDL-C, ApoB, and triglyceride values serve in the risk estimation in general populations as well as in people with diabetes. In patients with type 1 diabetes without diabetic nephropathy and HbA_1c <8.3% (67 mmol/mol) and women, LDL-C was a surprisingly poor predictor of CAD. Recent data suggest that in statin-treated patients, the strength of ApoB and non–HDL-C as risk predictors may be better than that of LDL-C (35). In this study, ApoB was a better predictor than non–HDL-C and LDL-C. However, the use of lipid-lowering agents (only a few patients were on fibrate therapy) at baseline was only 25.1 and 7.3% in patients with and without CAD, respectively. Furthermore, LDL-C was a good predictor of CAD in macroalbuminuric patients, in whom the use of lipid-lowering agents was much higher than in normoalbuminuric patients. Therefore, the use of lipid-lowering medication is an unlikely explanation for the poorer performance of LDL-C.

Patients with type 1 diabetes without microvascular complications have traditionally been considered to have a low risk of CAD; however, individuals with insulin resistance and other features of the metabolic syndrome should be recognized also from this patient group. The prevalence of double diabetes (i.e., when patients with type 1 diabetes show features of insulin resistance and type 2 diabetes) is increasing due to the societal trend of increased adiposity as well as weight gain caused by intensive glycemic control (27). Furthermore, the subcutaneous insulin administration leads to relative peripheral hyperinsulinemia and hepatic hypoinsulinemia, and a chronic adaption to this could reduce peripheral insulin uptake and increase hepatic glucose production, inducing insulin resistance (27). We have previously shown that in the FinnDiane population of type 1 diabetes, the overall prevalence of metabolic syndrome was 38% in men and 40% in women (36). In this study, ApoB and the ApoB/ApoA-I ratio were good predictors of CAD, but unfortunately their measurement generates additional costs and they are usually not available in the clinical setting. However, the triglyceride/HDL-C ratio, which can be calculated without any additional cost and strongly correlates with insulin resistance (37), was a good predictor of CAD in the patient groups in which LDL-C failed to predict a future CAD event.

In this study, the triglycerides predicted an incident CAD event in women, whereas in men, triglycerides were not a significant independent predictor. In the EURODIAB study, triglycerides also predicted CAD in women (22), and in the general population, triglycerides have been found to predict CVD more strongly in women (38). The current guidelines and the median values of the lipid variables in normoalbuminuric patients were close to each other, except regarding triglycerides, where the currently recommended level for triglycerides is <1.7 mmol/L. The median level for normo- and macroalbuminuric patients was 0.94 and 1.38 mmol/L, respectively, indicating increased lipolysis of triglyceride-rich lipoproteins due to activation of lipoprotein lipase by insulin therapy. Figure 1E illustrates how poorly the currently recommended cutoff level for triglycerides predicted a future CAD event in normoalbuminuric patients whereas the median level of 0.94 mmol/L performed much better. However, in macroalbuminuric patients, the current guideline level performed well. This observation may be due to the fact that atherogenic changes in LDL particle size and composition are seen even when triglyceride concentrations are <1.7 mmol/L (39). Serum triglyceride levels are strongly associated with glycemic control (11,40), but the results of all the figures were corrected for HbA_1c since it was included in the multivariate model. Furthermore, when patients were divided by the median HbA_1c level of the entire cohort, surprisingly triglycerides were a stronger predictor of CAD in patients with HbA_1c below the median than in patients with HbA_1c above. This suggests that there is an independent additive effect of triglycerides on CAD risk beyond that of glycemic control. Clustering of risk factors was also observed, and in a Cox regression analysis, an additive effect with the increasing number of risk factors was seen in patients with three or more risk factors.

Strengths of this study include the large cohort and that the lipid variables were all centrally measured. Patients were also geographically evenly distributed, resembling the distribution of the general population in Finland. Therefore, a selection bias is less likely than in single hospital-based studies. Furthermore, since the follow-up information was obtained from the HDR as well as the CDR, no patients were lost at follow-up. A limitation of the study is the possible survival bias; however, this was accounted for by performing competing risk regression analyses. Randomized clinical trials that take into account concomitant microvascular complications and with sufficiently long follow-up periods are needed to confirm our findings.

In conclusion, we have shown that in patients with type 1 diabetes, renal disease, glycemic control, and sex strongly modulated the relationship between lipid variables and CAD. Total and LDL-C were poor predictors of an incident CAD event in patients with normal AER and HbA_1c below the median of the cohort and women, in which the ratios and triglycerides performed better. However, the currently recommended cutoff level for triglycerides of 1.7 mmol/L may be too high to be able to predict an incident CAD event in these patients. Current treatment recommendations may need to be revised to capture residual CAD risk in patients with type 1 diabetes.

Acknowledgments. The authors acknowledge all the patients who participated in the study as well as all the physicians and nurses at the participating study centers (Supplementary Table 7). The authors are also thankful to H. Hildén, V. Naatti, and H. Perttunen-Mio (Division of Cardiology, Department of Medicine, Helsinki University Central Hospital) and M. Parkkonen, A.-R. Salonen, A. Sandelin, J. Tuomikangas, and T. Soppela (Folkhälsan Institute of Genetics, Folkhälsan Research Center, University of Helsinki) for their skilled technical assistance.

Funding. The study was supported by the Folkhälsan Research Foundation; the Wilhelm and Else Stockmann Foundation; the Finnish Medical Society (Finska Läkaresällskapet); the Finnish Kidney Foundation; the Biomedicum Helsinki Foundation; the Dorothea Olivia, Karl Walter, and Jarl Walter Perklen Foundation; the Novo Nordisk Foundation; the Liv och Hälsa Foundation; the Sigrid Jusélius Foundation; the Waldemar von Frenckell Foundation; and the American Heart Association (13POST17240095).

The funding sources were not involved in the design or conduct of the study in any way.

Duality of Interest. M.-R.T. has been a consultant for MSD, Kowa, Roche, Sanofi, and Novo Nordisk and has received travel and accommodation support during scientific meetings by MSD, Sanofi, and Novo Nordisk and lecture fees from AstraZeneca, MSD, Kowa, Sanofi, and Novo Nordisk. P.-H.G. has received research grants from Eli Lilly and Company and Roche, is an advisory board member for Boehringer Ingelheim, Abbott and AbbVie, Cebix, and Novartis, and has received lecture fees from Boehringer Ingelheim, Eli Lilly and Company, Genzyme, MSD, Novartis, Novo Nordisk, and Sanofi. No other potential conflicts of interest relevant to this article were reported.

The above-mentioned companies were not involved in the design or conduct of the study in any way.

Author Contributions. N.T. collected data, performed the statistical analyses, and wrote the manuscript. C.F. and V.-P.M. collected data, contributed to the discussion, and edited the manuscript. V.H., D.G., M.F., N.S., L.M.T., and J.W. collected data and edited the manuscript. M.-R.T. contributed to the discussion and edited the manuscript. P.-H.G. collected data, contributed to the discussion, and edited the manuscript. N.T. and P.-H.G. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.