Complementary prediction of cardiovascular events by estimated apo- and lipoprotein concentrations in the working age population. The Health 2000 Study

Abstract

Background. Apolipoprotein A-I (apoA-I) and B (apoB) and multiple lipoprotein cardiovascular risk factors can be computationally estimated with our extended Friedewald approach (EFW) from classical inputs. Their impact on cardiovascular events and mortality in the working age population is not known. Methods. The working age (≤ 65 years, n = 5956) prospective population-based cohort (follow-up of 7.8 ± 0.9 years; 46,572 patient years, 409 non-fatal incident cardiovascular events, and 55 cardiovascular and 266 all-cause deaths) had their total serum cholesterol (TC), triglycerides (TG), and HDL-C measured. Continuous net reclassification improvement (NRI) was calculated. Results. In Cox models adjusted with cardiovascular risk factors, EFW-HDL₂-C (HR 0.78, 95% CI 0.67–0.91; NRI 16.5%), apoA-I (HR 0.78, 95% CI 0.69–0.89; NRI 15.2%), apoB/apoA-I (HR 1.23, 95% CI 1.08–1.40; NRI 20.6%), and VLDL-TG (HR 1.15, 95% CI 1.05–1.25; NRI 20.1%) were associated with incident non-fatal cardiovascular events and improved risk prediction compared with TC, LDL-C, or non-HDL-C. Cardiovascular deaths could be best predicted with EFW apoB (HR 1.81, 95% CI 1.18–2.77; NRI 77.3%). Conclusions. EFW approach-derived HDL₂-C, apoA-I, apoB/apoA-I, and VLDL-TG improve prediction of non-fatal cardiovascular events, and apoB of cardiovascular mortality, and can be utilized for risk estimation in a working age population without extra cost.

Keywords::

• Apolipoproteins
• cardiovascular
• computational methods
• events
• extended
• Friedewald
• mortality
• neural networks

Key messages

• Apolipoprotein A-I (apoA-I) and B (apoB) and multiple lipoprotein risk factors for vascular diseases can be accurately estimated by our newly developed extended Friedewald approach (EFW) from classical Friedewald inputs.

• EFW-estimated HDL₂-C, apoA-I, apoB/apoA-I, and VLDL-TG were significantly associated with incident non-fatal cardiovascular events and improved risk prediction over TC, LDL-C, and non-HDL-C. Cardiovascular deaths could be best predicted using EFW-estimated apoB.

• The EFW approach, without extra cost, provides predictive value for non-fatal cardiovascular events and cardiovascular mortality and can be utilized for risk estimation in clinical practice at the population level in a working age population.

Introduction

Low-density lipoprotein cholesterol (LDL-C), currently most often estimated via the Friedewald formula (1), is considered the major atherogenic lipoprotein measure and is routinely determined in epidemiological studies and also guides clinicians in clinical practice (2). According to a recent large-scale study, the clinical utility of direct LDL-C assays is questionable (3). Actually, intermediate-density lipoprotein cholesterol (IDL-C) relates to the progression of coronary and carotid artery atherosclerosis (4,5), but it is not obtained during routine lipid measurements. In contrast, high-density lipoprotein cholesterol (HDL-C) is inversely associated with the risk for vascular complications and considered as an anti-atherogenic lipoprotein (6,7). More precisely, the HDL₂ subfraction, also undeterminable by routine methods, is a more powerful protector than HDL₃ (8).

Accumulating evidence suggests that lipoprotein and apolipoprotein (apo) measures could be better markers of atherosclerosis than classical cholesterol parameters and could be used to improve risk stratification of vascular complications (9–12). Since apoB is the major protein in the LDL particles and apoB concentration reflects the total number of atherogenic apoB-containing lipoprotein particles, it can surpass LDL-C and non-HDL-C in cardiovascular risk assessment (11,13–15). Conversely, apoA-I is the major protein in the HDL particles and important in the removal of excess cholesterol in reverse cholesterol transport (16,17). ApoB also reflects the concentration of atherogenic triglyceride (TG)-rich lipoprotein particles and in combination with apoA-I indicates the increased risk in metabolic syndrome, i.e. increased TG and decreased HDL.

It has been proposed that apoB and the apoB/apoA-I ratio should be determined for risk stratification (13). Actually, the non-fasting apoB/apoA-I ratio could be used as a simple indicator of cardiovascular risk, and it seems superior to any of the cholesterol ratios for estimation of the risk of acute myocardial infarction (12).

The classical Friedewald formula (FW) requires data on serum TG, serum cholesterol (TC), and HDL-C (1). However, it is valid only if serum TG ≤ 4.52 mmol/L (18). In addition, the estimated values do not represent pure LDL-C but include also a contribution of the intermediate-density lipoprotein (IDL). The new extended Friedewald approach (EFW) based on artificial neural network regression algorithms can computationally yield estimates of the apoB and apoA-I concentrations, VLDL-TG, IDL-C, LDL-C, and HDL₂-C without the laborious determinations requiring ultracentrifugation by utilizing data on classical FW inputs. The technology is based on the ability of neural networks to learn inherent relations if appropriate training data are available. We have utilized patient cohorts with data on both classical FW parameters and actual measured parameters for the development of EFW. Actually apoB, apoB/apoA-I, and IDL-C provided better prediction of mortality in type 1 diabetics than the conventional FW parameters (19). Thus, apart from type 1 diabetes, at the population level, and in large epidemiological studies in working age subjects, the impact of the anti-atherogenic (HDL₂) and the impact of EFW estimates of pro-atherogenic lipoprotein particles on non-fatal cardiovascular events and survival is not known.

In the present study, we utilized a large working age population cohort and analyzed the effect and value of the EFW measures in the prediction of incident non-fatal cardiovascular events and long-term cardiovascular mortality. We hypothesized that EFW parameters could be utilized to predict non-fatal cardiovascular events and survival at the population level.

Material and methods

Health 2000 Study cohort description

This study is a part of the Health 2000 Survey, a Finnish population-based study, carried out in 2000–2001 and having a representative stratified random cluster sample of the Finnish population. The Health 2000 Study sample of working age subjects (30–65 years) comprised 5956 subjects. Following a home interview, a comprehensive health examination including questionnaires, measurements (e.g. blood pressure, resting ECG), and physician's physical examination was performed. The implementation of the survey is described in detail elsewhere (20). Of the 5956 subjects, 241 (4.0%) reported use of lipid-lowering medication, and 4689 (78.7%) reported no use of these drugs. Data were missing for 1026 (17.2%) subjects. The Health 2000 Survey obtained contemporary information about major diseases in Finland. The National Hospital Discharge Register and the national register on rights to reimbursements for medication costs were linked to the Health 2000 Survey data. The study protocol of the Health 2000 Survey was approved by the Epidemiology Ethics Committee of the Helsinki and Uusimaa Hospital District. The participants in the survey signed an informed consent both before the health interview and at the beginning of the health examination.

Laboratory tests, estimation of lipoprotein lipid, and apoA-I and apoB concentrations

Venous blood samples were drawn from the antecubital vein. Serum TG, TC, HDL-C, and glucose concentrations were determined enzymatically (Roche Diagnostics, GmbH, Mannheim, Germany for HDL; Olympus System Reagent, Hamburg, Germany for total cholesterol, triglycerides, and glucose) with a clinical chemistry analyzer (Olympus, AU400, Hamburg, Germany). LDL cholesterol was calculated with the Friedewald formula (1). ApoA-I and apoB concentrations were determined immunoturbidimetrically. For the estimation of VLDL-TG, IDL-C, LDL-C, and HDL₂ as well as apoA-I and apoB concentrations from the measured inputs, i.e. TG, TC, and HDL-C, we used a newly developed and validated extended Friedewald approach (EFW) freely available online (http://www.computationalmedicine.fi/software/Lipido) (19). The LDL cholesterol estimated via the traditional Friedewald formula (abbreviated here as LDL-C) includes also the IDL-C, while the corresponding measure from the extended Friedewald approach (EFW-LDL-C) is referring to pure LDL, i.e. they represent fundamentally different measures as detailed in (18).

Other measurements and definitions

Clinical hypertension was defined as a clinic BP ≥ 140/90 mmHg and/or the use of antihypertensive medication. Diabetes mellitus was defined as a serum glucose level of 7.0 mmol/L or greater or a history of the use of oral hypoglycemic agents or insulin injections. Smoking was defined as daily use of tobacco products.

Follow-up, end-points, survival, and causes of death

The follow-up death data were received from the Causes of Death Register, maintained by Statistics Finland (http://www.stat.fi/), at the beginning of June 2009; this source has been shown to be reliable (21). Subsequently, the follow-up data on incident non-fatal cardiovascular events were obtained from the national database (The Finnish Hospital Discharge Register). Incident non-fatal cardiovascular events included were specified according to ICD-10 (International Classification of Diseases) and included any cardiac, brain-related (amaurosis fugax, transient ischemic attack, ischemic stroke), and peripheral events (ICD-10 codes: G45.0–G46.8, I11.0–I74.9) and procedures (angiography, percutaneous endovascular procedures, and arterial bypass procedures) associated with atherosclerosis. During the follow-up, of the 5956 patients, 409 (6.9%) had incident non-fatal cardiovascular events, of which 112 (1.9%) were brain-related. There were 58 (1.0%) percutaneous endovascular procedures and 77 (1.3%) arterial bypass procedures.

Of the 5956 patients, 266 (4.5%) died of all causes and 55 (0.9%) of cardiovascular causes during the follow-up. The causes of death based on ICD-10 classifications were also obtained and divided further into cardiovascular (n = 55, 20.7%; cardiac n = 46, brain-related i.e. ischemic stroke, bleeding, vascular dementia n = 8, and other vascular causes n = 1), and other causes (n = 8), i.e. cancer, infection, trauma, unknown. The mean follow-up was 7.8 ± 0.9 years, and the cumulative follow-up was 46,572 patient years.

Statistical analysis and power calculations

Power calculation for different lipoprotein measurements was performed. Given the standard deviations of lipoprotein measurement data from the cohort, a power level of > 90% was achieved to reach clinically meaningful relative risk of 1.4 or more of end-points (incident non-fatal cardiovascular events and cardiovascular mortality) with two-sided alpha of 0.05 for all lipoprotein classes. Power calculations were made with free Power and Sample Size Calculation program (available from: http://biostat.mc.vanderbilt.edu/twiki/bin/view/Main/PowerSampleSize). The agreement between measured and EFW estimates of apoA-I, apoB, and apoB/apoA-I was examined by Pearson correlation coefficients (R) and checked with scatter plots for outliers and non-linear behavior and Bland–Altman plots by plotting the difference between estimated and measured concentrations (bias) and the mean of measured and estimated concentrations (Figure 1). The extended Friedewald estimates of apoA-I, apoB, and apoB/apoA-I had strong correlation with actual measured concentrations (Pearson R = 0.91, P < 0.0001; R = 0.96, P < 0.0001; and R = 0.94, P < 0.0001, respectively). As indicated by the Bland– Altman plots, the agreement between the estimated and measured values for apoB and apoA-I as well as for their calculated ratio was good with only very small biases. The mean bias was –0.0260 g/L for apoA-I, 0.0072 g/L for apoB, and 0.0110 for the apoB/apoA-I ratio. The effect of gender on various parameters was analyzed using Student's t test for independent samples assuming unequal variances.

Figure 1. Correlation of (A) apoA-I, (B) apoB, and (C) apoB/apoA-I extended Friedewald estimates with actual measured concentrations. The dashed lines represent regression lines. Bland–Altman plots for the cross-validation of measured and extended Friedewald estimates of (D) apoA-I, (E) apoB, and (F) apoB/apoA-I. Bias was determined as difference between estimated and measured values. The dashed lines represent regression lines. The solid lines represent mean bias ± SD. The dashed line represents the mean bias and the solid lines represent ± 2 SD.

Cox regression proportional hazards analysis was performed. The analyses were first adjusted according to age and gender (model 1). Further analysis was done adjusting with age, gender, hypertension, diabetes, current smoking, and lipid-lowering medication (model 2). Variables were scaled to zero mean and unit SD before Cox regression analysis. Data are presented as standardized hazard ratios per one SD increase. To evaluate the incremental predictive value of different lipoprotein (either estimated or measured) parameters for cardiovascular outcomes over classical cardiovascular risk factors, continuous net reclassification improvement (NRI) (22,23), C-statistic, and Hosmer– Lemeshow chi-square were calculated. NRI is presented as net proportion (%) of events with increased model-based probability plus net proportion of non-events with decreased model-based probability. The analyses were done adjusting for the traditional risk factors. First, multivariable logistic regression models were generated with the following variables: age, sex, hypertension, diabetes, current smoking, and lipid-lowering medication (model 2) and a measured, FW-estimated, or EFW-estimated parameter. Then, predicted risks for all individuals were calculated using the R software function predRisk() provided in the PredictABEL R package (24). Finally, continuous NRI was evaluated using improveProb() routine (25) in R software to test whether predictions from model 2 are significantly different from predictions from model 2 plus a measured, FW-estimated, or EFW-estimated parameter. Framingham risk scores (FRS) for 10-year cardiovascular risk were determined as described previously (26), and log-transformed FRS was used to recalibrate the score for the present study (model 3). Continuous NRI was evaluated similarly to test whether predictions from model 3 are significantly different from model 3 plus a measured, FW-estimated, or EFW-estimated parameter. Statistical analyses were performed with the SPSS release 14.0 for Windows (SPSS Inc., Chicago, IL, USA) and the R statistical software version 2.13.1 (http://www.r-project.org). P < 0.05 was considered statistically significant in all the analyses.

Results

Characteristics of study population

The characteristics of serum lipid and apolipoprotein concentrations are detailed in Table I. Of the 5956 patients, 3023 (50.8%) were women, 1454 (26.5%) had hypertension, 210 (3.8%) had diabetes, and 1461 (26.6%) were regular smokers. The mean age was 46.3 ± 9.7 (30–65) years. The concentrations of pro-atherogenic factors (TG, TC, LDL-C, LDL-C/HDL-C, TC/HDL-C, non-HDL-C, apoB, apoB/apoA-I, VLDL-TG, IDL-C, EFW-LDL-C) were significantly higher in men compared with women, while anti-atherogenic factors (HDL-C, apoA-I, HDL₂-C, HDL₃-C) were higher in women (Table I).

Table I. Characteristics of the measured and estimated serum lipoprotein and apolipoprotein concentrations in the cohort. Data are presented as mean (SD). FW refers to the Friedewald formula (1) and EFW to the extended Friedewald approach (19).

	All	Men	Women	P value
Serum/FW-estimated (mmol/L)
TG	1.57 (1.09)	1.84 (1.34)	1.34 (0.71)	< 0.0001
TC	5.90 (1.11)	5.99 (1.12)	5.81 (1.09)	< 0.0001
HDL-C	1.34 (0.38)	1.21 (0.34)	1.46 (0.38)	< 0.0001
LDL-C	3.68 (1.05)	3.85 (1.06)	3.53 (1.02)	< 0.0001
LDL-C/HDL-C	3.00 (1.32)	3.44 (1.41)	2.61 (1.09)	< 0.0001
TC/HDL-C	4.76 (1.70)	5.33 (1.87)	4.24 (1.33)	< 0.0001
non-HDL-C	4.56 (1.15)	4.78 (1.17)	4.36 (1.10)	< 0.0001
apoA-I (g/L)	1.60 (0.29)	1.52 (0.26)	1.67 (0.30)	< 0.0001
apoB (g/L)	1.21 (0.30)	1.28 (0.30)	1.14 (0.28)	< 0.0001
apoB/apoA-I	0.78 (0.25)	0.87 (0.26)	0.70 (0.21)	< 0.0001
EFW-estimated (mmol/L)
VLDL-TG	0.90 (0.78)	1.09 (0.94)	0.72 (0.54)	< 0.0001
IDL-C	0.33 (0.15)	0.36 (1.59)	0.30 (0.14)	< 0.0001
EFW-LDL-C	3.87 (0.89)	3.93 (0.90)	3.84 (0.87)	0.017
HDL2-C	0.85 (0.35)	0.73 (0.30)	0.95 (0.36)	< 0.0001
HDL3-C	0.50 (0.05)	0.49 (0.06)	0.51 (0.05)	< 0.0001
apoA-I (g/L)	1.57 (0.25)	1.50 (0.22)	1.63 (0.25)	< 0.0001
apoB (g/L)	1.21 (0.27)	1.28 (0.28)	1.15 (0.26)	< 0.0001
apoB/apoA-1	0.79 (0.22)	0.87 (0.21)	0.72 (0.20)	< 0.0001

Incident non-fatal cardiovascular events

Of the measured and FW-estimated parameters, HDL-C, apoA-I, apoB/apoA-I, TC/HDL-C, LDL-C/HDL-C, TG, and apoB were significantly associated with incident non-fatal cardiovascular events both after adjustment for age and gender (model 1) and further adjustments for hypertension, diabetes, current smoking, and lipid-lowering medication usage (model 2) (Figure 2). Interestingly, non-HDL-C, TC, and LDL-C showed no association at all.

Figure 2. Ranked association of measured and Friedewald (FW)-estimated parameters and extended Friedewald (EFW) estimates with (A) incident non-fatal cardiovascular events (brain, cardiac, peripheral vascular event/procedure), (B) cardiovascular mortality according to Cox regression proportional hazards survival analysis adjusted with age, gender (model 1: upper panels) and age, gender, hypertension, diabetes, current smoking, and lipid-lowering medication (model 2: lower panels). Variables were scaled to zero mean and unit SD before Cox regression analysis. Data are presented as standardized hazard ratios per 1 SD increase.

Of the EFW estimates HDL₂-C, apoA-I, HDL₃-C, apoB/apoA-I, and VLDL-TG were significantly associated with incident non-fatal cardiovascular events in both models 1 and 2, while EFW-LDL-C showed significant association only in model 1 (Figure 2). The values for EFW-estimated and measured apoA-I, apoB, and apoB/apoA-I were of the same magnitude.

Cardiovascular mortality

Of the measured and FW-estimated parameters, apoB, HDL-C, LDL-C/HDL-C, non-HDL-C, apoB/apoA-I, and TC/HDL-C were significantly associated with cardiovascular mortality in both models 1 and 2 (Figure 2).

Of the EFW estimates HDL₂-C, apoB, apoB/apoA-I, and IDL-C showed significant association with cardiovascular mortality in both models 1 and 2 (Figure 2).

Value of extended Friedewald parameters in the prediction of non-fatal incident cardiovascular events and mortality compared with classical parameters

In the prediction of incident non-fatal cardiovascular events, in adjusted analyses (model 2), of the measured and FW-estimated parameters apoB/apoA-I (NRI 26.0%), TC/HDL-C (NRI 25.4%), apoB (NRI 24.9%), apoA-I (NRI 24.1%), LDL-C/HDL-C (NRI 23.5%), TG (NRI 16.8%), HDL-C (NRI 15.1%) and of the EFW estimates, apoB (NRI 25.2%), IDL-C (NRI 23.0%), apoB/apoA-I (NRI 20.6%), VLDL-TG (NRI 20.1%), HDL₂-C (NRI 16.5%), and apoA-I (NRI 15.2%) showed statistically significant NRI (Figure 3). The corresponding C-statistic and Hosmer–Lemeshow chi-square (H-L χ²) values are shown in Table II.

Figure 3. Ranked net reclassification improvement (NRI) of measured and Friedewald (FW)-estimated parameters and extended Friedewald (EFW) estimates with (A) incident non-fatal cardiovascular events (brain, cardiac, peripheral vascular event/procedure), (B) cardiovascular mortality relative to standard model adjusted with age, gender, hypertension, diabetes, current smoking, and lipid-lowering medication (model 2).

Table II. Ranked net reclassification improvement (NRI) of measured and Friedewald (FW)-estimated parameters and extended Friedewald (EFW) estimates with incident non-fatal cardiovascular events (brain, cardiac, peripheral vascular event/procedure). Cox regression proportional hazards survival analysis adjusted for age, gender, hypertension, diabetes, current smoking, and lipid-lowering medication. C-statistic, Hosmer–Lemeshow chi-square (H-L χ²).

Incident cardiovascular events	C-stat	H-L χ^2	NRI (95% CI) (%)
FW/Established risk factors (model 2) added with:	0.831	18.3	reference
apoB/apoA-I	0.838	9.8	26.0 (12.0, 39.9)
TC/HDL-C	0.837	14.0	25.4 (11.5, 39.4)
apoB	0.836	9.1	24.9 (10.9, 38.8)
apoA-I	0.839	15.4	24.1 (10.2, 38.1)
LDL-C/HDL-C	0.838	13.3	23.5 (9.5, 37.5)
TG	0.835	9.8	16.8 (2.9, 30.8)
HDL-C	0.837	13.2	15.1 (1.2, 29.1)
non-HDL-C	0.835	6.4	11.2 (–2.7, 25.2)
LDL-C	0.836	9.5	7.2 (–6.7, 21.2)
TC	0.835	14.4	4.7 (–9.2, 18.6)
EFW-estimated established risk factors (model 2) added with:	0.831	18.3	reference
apoB	0.837	11.8	25.2 (10.9, 39.5)
IDL-C	0.836	10.1	23.0 (9.0, 36.9)
apoB/apoA-I	0.839	14.0	20.6 (6.3, 35.0)
VLDL-TG	0.835	8.5	20.1 (6.1, 34.0)
HDL2-C	0.833	11.5	16.5 (2.2, 30.8)
apoA-I	0.840	16.0	15.2 (0.9, 29.5)
HDL3-C	0.833	15.8	6.3 (–8.0, 20.6)
EFW-LDL-C	0.835	11.4	1.3 (–12.7, 15.2)

In the prediction of cardiovascular mortality, in adjusted analyses (model 2), of the measured and FW-estimated parameters, non-HDL-C (NRI 74.9%), apoB/apoA-I (NRI 63.5%), TC/HDL-C (NRI 60.8%), TC (NRI 60.0%), LDL-C/HDL-C (NRI 59.4%), apoB (NRI 57.4%), LDL-C (NRI 55.2%) and of the EFW estimates, apoB (NRI 77.3%), IDL-C (NRI 56.9%), EFW-LDL-C (NRI 51.3%), and apoB/apoA-I (NRI 50.7%) showed statistically significant NRI (Figure 3). The corresponding C-statistic and Hosmer–Lemeshow chi-square (H-L χ²) values are shown in Table III.

Table III. Ranked association net reclassification improvement (NRI) of measured and Friedewald (FW)-estimated parameters and extended Friedewald (EFW) estimates with cardiovascular mortality. Cox regression proportional hazards survival analysis adjusted for age, gender, hypertension, diabetes, current smoking, and lipid-lowering medication. Variables were scaled to zero mean and unit SD before Cox regression analysis. C-statistic, Hosmer–Lemeshow chi-square (H-L χ²). Infinite value (∞) due to low number of cases.

Cardiovascular mortality	C-stat	H-L χ^2	NRI (95% CI) (%)
FW/Established risk factors (model 2) added with:	0.878	∞	reference
non-HDL-C	0.887	∞	74.9 (39.6, 110.2)
apoB/apoA-I	0.884	∞	63.5 (28.2, 98.8)
TC/HDL-C	0.883	∞	60.8 (25.5, 96.1)
TC	0.884	∞	60.0 (24.6, 95.3)
LDL-C/HDL-C	0.881	∞	59.4 (24.1, 94.7)
apoB	0.893	∞	57.4 (22.1, 92.7)
LDL-C	0.882	∞	55.2 (19.9, 90.5)
TG	0.877	∞	16.0 (–19.3, 51.3)
HDL-C	0.874	∞	10.8 (–24.6, 46.1)
apoA-I	0.872	∞	1.0 (–34.3, 36.3)
EFW-estimated established risk factors (model 2) added with:	0.878	∞	reference
apoB	0.890	∞	25.2 (10.9, 39.5)
IDL-C	0.881	∞	23.0 (9.0, 36.9)
EFW-LDL-C	0.876	∞	20.6 (6.3, 35.0)
apoB/apoA-I	0.880	∞	20.1 (6.1, 34.0)
VLDL-TG	0.878	∞	16.5 (2.2, 30.8)
apoA-I	0.872	∞	15.2 (0.9, 29.5)
HDL2-C	0.878	∞	6.3 (–8.0, 20.6)
HDL3-C	0.879	∞	1.3 (–12.7, 15.2)

Additive value of extended Friedewald parameters in the prediction of non-fatal incident cardiovascular events and mortality compared with Framingham score

In adjusted analyses (model 3), of the measured and FW-estimated parameters, TC/HDL-C (NRI 31.4%), TG (NRI 28.8%), apoB/apoA-I (NRI 15.2%) and of the EFW estimates VLDL-TG (NRI 27.3%) had statistically significant NRI indicating additive value to Framingham score in the prediction of non-fatal incident cardiovascular events (Figure 4).

Figure 4. Ranked net reclassification improvement (NRI) of measured and Friedewald (FW)-estimated parameters and extended Friedewald (EFW) estimates with (A) incident non-fatal cardiovascular events (brain, cardiac, peripheral vascular event/procedure), and (B) cardiovascular mortality relative to standard model based on Framingham risk score (26) for 10-year cardiovascular risk (model 3).

In adjusted analyses (model 3), of the measured and FW- estimated parameters, TG (NRI 72.3%) and of the EFW estimates, EFW-LDL-C (NRI 54.9%) and HDL₃-C (NRI 40.9%) had statistically significant NRI indicating additive value to Framingham score in the prediction of cardiovascular mortality (Figure 4), while VLDL-TG showed negative NRI.

Discussion

Selected EFW outputs, HDL₂-C, apoA-I, apoB/apoA-I, and VLDL-TG are promising candidates for the assessment of the risk of incident non-fatal cardiovascular events in a working age population, improving risk prediction compared with TC, LDL-C, or non-HDL-C. Of these, VLDL-TG even improved significantly the predictive value of Framingham risk score. Their value in the prediction of cardiovascular mortality remains to be elucidated in larger cohorts, although EFW apoB ranked first in the prediction of cardiovascular mortality. The present method is free of additional costs, and the estimates can be obtained from existing cohorts by implementing the algorithm available online (http://www.computationalmedicine.fi/software/Lipido) (19).

When focusing on the triglyceride-rich lipid fractions, although VLDL-TG showed association with incident non-fatal cardiovascular events, no association was found with cardiovascular mortality. As a marker of an intermediate between the TG-rich VLDL and C-rich LDL particles, EFW-derived IDL-C remained a significant and independent predictor of cardiovascular mortality after adjustment for age, gender, hypertension, diabetes, use of lipid-lowering drugs, and current smoking. It also provided additional value to prediction of both incident events and mortality compared with classical risk factors. This finding is in line with previous studies in which IDL-C was strongly associated with the progression of atherosclerosis (4,5) and mortality in type 1 diabetics (HR 1.45) (19) and in non-diabetic subjects (4,5).

Our findings of only weak associations of baseline LDL-C with cardiovascular events and cardiovascular mortality are in line with previous studies (2,4,5). The problem is that LDL-C concentration varies significantly between individuals with the same LDL particle size (27,28). Respectively, LDL-C does not necessarily reflect LDL particle size since metabolic reactions alter both lipoprotein size and composition. In addition, the relative amounts of cholesterol and triglycerides in LDL particles vary significantly between individuals (29). Furthermore, LDL-C cannot accurately reflect changes related to obesity and type 2 diabetes (30). It must also be taken into account that the classical FW formula is valid only if serum TG ≤ 4.52 mmol/L. In addition, the estimated values do not represent pure LDL-C but include also a contribution from IDL. Taken together, our results are in line with previous observations that the triglyceride pathway (TG, VLDL, IDL) is associated with atherosclerotic vascular disease (31,32).

Our results of the strong protective effect of HDL-C, HDL₂-C, and HDL₃-C on incident non-fatal cardiovascular events and of HDL-C and HDL₂-C on cardiovascular mortality are in line with previous studies (6,7,33). In the present study, despite a similar effect on cardiovascular events, EFW-derived HDL₂-C was observed to be a stronger predictor of incident cardiovascular events and improved net reclassification more than HDL-C or HDL₃-C. Our results are in line with a previous study (8), and this underlines the importance of analyzing the effect of different HDL subclasses on cardiovascular events. However, this result must be interpreted with caution since routine clinical measurements of serum cholesterol primarily reflect levels of large, cholesterol-rich particles and may lack the sensitivity to detect small, cholesterol-poor HDL (34). The extended Friedewald approach provides an efficient method to determine the sole effect of HDL₂-C in existing cohorts.

Our results of the protective effect of apoA-I and the detrimental effect of apoB/apoA-I or apoB on incident non-fatal cardiovascular events are supported by previous data in which it has been demonstrated that they are better markers of cardiovascular mortality than serum lipids (9–11,31,35,36). ApoA-I, apoB/apoA-I, and apoB also improved risk prediction of incident cardiovascular events over standard models based on TC, LDL-C, non-HDL-C and cardiovascular risk factors. ApoB/apoA-I and apoA-I also predict the development of subclinical atherosclerosis better than serum lipids, i.e. they are associated with carotid intimal-media thickness and impaired brachial artery endothelial function (37). Our findings are similar to the results of a previously published large case-control study (INTERHEART) reporting strong association between apolipoproteins and the incidence of myocardial infarction (MI) (12). More precisely, the measured apoB/apoA-I ratio was linked with the highest population-attributable risk for MI when compared to other lipid fractions. In the INTERHEART study, a 1-SD change in measured apoB/apoA-I corresponded to a 59% increase in the risk for MI (12), whereas we found that a 1-SD change in estimated apoB/apoA-I corresponded to a 23% increase in incident non-fatal events and a 65% increase in the risk for CV death. One previous study (based on a multi-ethnic US population) confirms our results regarding the risk of CV death, with even higher risk increases associating with apoB/apoA-I (38). Similarly, apoB has been shown to be a significant predictor of CHD and mortality due to CHD (11,39). The protective effect of apoA-I against incident non-fatal cardiovascular events, observed in the present study, is also in line with previous findings on the role of apoA-I in the protection against development of atherosclerosis (40). However, the correlation between the measured and estimated apoA-I was weaker than with apoB. This is probably due to the fact that LDL particles contain a single copy of apoB and there are two to five apoA-I molecules in a single HDL₂ and HDL₃ particle (41). Currently, apoB and apoA-I are not recommended for routine use due to additional expenses and laborious laboratory processes, and due to the fact that therapeutic cut-off points have not been defined. At present, measurement of apoB and apoA-I has been included in guidelines for diabetic subjects (42). Our results encourage reanalysis of existing cohorts to overcome these issues. The EFW approach was validated in type I diabetics where apoB and apoB/apoA-I were demonstrated to associate with overall mortality (19).

Although the follow-up period was long (over 7 years), the number of CV deaths in the present population was fairly small, limiting the power of the study. The relatively low mortality rate can be attributed to the fact that this study was based on a healthy general population ≤ 65 years of age. However, thanks to well documented cardiovascular risk factors, we were able to get reasonably accurate results in the survival analyses. However, as serum apoB and apoA-I as well as their calculated apoB/apoA-I ratio have been shown to be important risk predictors also among older people (over 70 years of age), given a sufficiently large study population, it is probable that EFW parameters could enhance risk prediction accuracy also in older age-groups (11). The causes of death are based on national registry data, which is not explicitly reliable, while dates of death have been obtained in all the patients. However, the median age of the population is 46, therefore the majority of patients are at risk for atherosclerotic events. The validity of the EFW parameters has previously been demonstrated (19), and our present validation data strongly support these observations. Net reclassification improvement (NRI) was used to analyze the predictive value since the C-statistic is rather insensitive in detecting clinically meaningful changes when parameters are added to a set of already good predictors (22). This was reflected by extremely small changes in C-statistic values.

At present, determination of HDL₂-C, apoA-I, apoB/apoA-I, apoB, and VLDL-TG has not been included in cardiovascular guidelines for lipid management, except in Canadian guidelines which include target apoB as < 0.8 g/L (43). Our results demonstrate that complementary EFW parameters, HDL₂-C, apoA-I, apoB/apoA-I, and VLDL-TG, derived from conventional Friedewald inputs without any additional cost, are independent predictors of non-fatal cardiovascular events at the population level in working age subjects and provide additional value compared with models based either on non-HDL, LDL-C, or TC. Interestingly, EFW apoB showed superior predictive value in the prediction of cardiovascular deaths.

Acknowledgements

We thank the personnel in the field and the supporting organizations of the Health 2000 Survey. Financial support was received from the Finnish Cultural Foundation, the Finnish Foundation for Cardiovascular research (M.A.-K., T.L.), from the Academy of Finland Responding to Public Health Challenges Research Programme (M.A.-K.), and the Tampere University Hospital Medical Fund (EVO grant 9M048 for T.L.).

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.