Residual cardiovascular risk in individuals on lipid-lowering treatment: quantifying absolute and relative risk in the community

Study sample

The present analyses were based on data from attendees at the sixth examination cycle of the Framingham Offspring cohort (1995–1998).8 A total of 520 individuals were excluded for reasons detailed in the online supplementary material, resulting in a sample size of n=3012. For 95 individuals, information regarding the subclinical disease burden of the carotid artery (detailed below) was missing, resulting in a sample size of n=2917 for the analyses related to subclinical carotid disease burden. Circulating C reactive protein (CRP) concentrations and plasma levels of B-type natriuretic peptide (BNP) were available in 2811 participants. All participants provided written informed consent.

Laboratory measurements

Lipid traits, including blood levels of total cholesterol, high-density lipoprotein (HDL) cholesterol and triglycerides, were measured in fasting participants using standardised assays. LDL-C was calculated using the Friedewald formula in individuals with triglyceride levels <400 mg/dL.9 Measurements of CRP and BNP are described in the online supplementary material.

Ultrasound of the carotid artery

Ultrasonographic measurements of the carotid arteries on both sides were performed using a 7.5 MHz transducer and a 5.0 MHz transducer for images of the common and of the internal carotid arteries, respectively, on a Toshiba SSH-140 A machine, following a standardised protocol.10–12 More details regarding the ultrasound measurements are provided in the online supplementary material.

‘Carotid ultrasound abnormality’ was defined as (1) increased (≥80^th sex-specific percentile) intima media thickness (IMT), a combined standardised measure including information from the internal and common carotid arteries; (2) an extreme increase of the common carotid IMT ≥1 mm; or (3) significant stenosis (≥25% narrowing) of the common or internal carotid arteries, consistent with prior publications.13 14 We chose this combined parameter in order to reflect both alterations in IMT (a measure of overall atherosclerotic plaque burden) and the presence of arterial stenoses (a measure of obstructive atherosclerosis).

Adjudication of incident CVD events

Framingham participants are under regular surveillance for new-onset CVD events. More details regarding the adjudication of events are provided in the online supplementary material.

Statistical analyses

Based on the intake of lipid-lowering medications and the blood LDL levels determined at the sixth examination cycle, participants were assigned to one of five mutually exclusive groups.

1. usual LDL-C <100 mg/dL, not on LLT

2. usual LDL-C ≥100 mg/dL to <130 mg/dL, not on LLT

3. treated LDL-C <130 mg/dL on LLT

4. usual LDL-C ≥130 mg/dL, not on LLT

5. treated LDL-C ≥130 mg/dL on LLT.

There were very few individuals (n=58) with LDL-C <100 mg/dL on LLT; therefore no separate category was created for these individuals. We chose these cut points taking into account different target lipid levels and the presence versus absence of LLT, thereby ensuring that the strata yielded an adequate number of events to enable analyses with reasonable statistical power.

To quantify the baseline risks in the different LDL groups, we calculated the 10-year probability of a CVD event15 for each participant and displayed these event probabilities as boxplots according to LDL group. Treated individuals (groups 3 and 5) were compared versus untreated individuals (groups 1, 2 and 4) with respect to their 10-year CVD event probabilities using analysis of variance.

Subclinical disease burden at the carotid arteries by LDL group

First, we calculated the ORs for the binary trait ‘carotid ultrasound abnormality’ (defined as detailed above, including information on IMT and on stenoses) for each LDL group, using group 1 as the referent, in age-and sex-adjusted as well as multivariable-adjusted logistic regression models. The multivariable models included age, sex, systolic blood pressure, antihypertensive medication, smoking and diabetes mellitus. Second, we compared the odds for ‘carotid ultrasound abnormality’ in individuals on LLT (group 5 and group 3) with the odds for ‘carotid ultrasound abnormality’ in untreated individuals with similar LDL levels (group 4 and groups 1+2 combined, respectively) using age- and sex-adjusted as well as multivariable-adjusted logistic regression models. We focused on ‘carotid ultrasound abnormality’ as the primary subclinical disease trait because subclinical alterations of the carotid artery are reported to be strongly associated with increased LDL levels16 and the selected trait ‘carotid ultrasound abnormality’ includes both information from IMT and of relevant stenoses of the carotid arteries.13 14

Association of LDL category with incident CVD

We performed several analyses to describe the residual cardiovascular risk associated with different levels of LDL-C, stratified by LLT status, age group and sex. First, the unadjusted survival free of CVD by ‘LDL category’ was graphically displayed using a Kaplan-Meier curve. Second, we report absolute disease risks (events per 1000 person-years) for each LDL category for the entire sample, and stratified by sex and by age group (<60 and ≥60 years). Third, Cox proportional hazard models were used to relate each LDL group to incident CVD, thereby adjusting for age, sex, systolic blood pressure, antihypertensive medication, smoking and diabetes mellitus. In secondary analyses, we additionally adjusted for presence versus absence of carotid ultrasound abnormality (defined as above) and for circulating CRP and BNP levels. In these models, we compared LDL categories 2–5 (each category considered separately) with LDL category 1 (usual LDL-C <100 mg/dL, no LLT). Fourth, we specifically compared treated with untreated individuals with similar LDL levels. Thus, we calculated multivariable-adjusted HR for incident CVD for group 5 (using group 4 as the referent) and for group 3 (using groups 1+2 combined as the referent).

Association of LDL group with subclinical disease in the carotid arteries

In age- and sex-adjusted and in multivariable-adjusted models, individuals on lipid-lowering medication (groups 3 and 5) and individuals with an untreated LDL above 130 mg/dL (group 4) had significantly higher ORs for carotid ultrasound abnormalities as compared with individuals in the referent group (usual LDL <100 mg/dL; figure 1A,B). Furthermore, individuals on treatment who reach LDL levels <130 mg/dL (group 3) had statistically significantly increased odds for subclinical carotid ultrasound abnormalities as compared with untreated individuals with similar LDL levels (groups 1+2 combined; OR: 1.95, 95% CI 1.34 to 2.84, P=0.0005). No statistically significant differences between treated individuals with LDL ≥130 (group 5) and untreated individuals with comparable LDL levels (group 4; OR: 1.04, 95% CI 0.69 to 1.57, P=0.86) could be observed.

Request permission

Residual risk for incident CVD in individuals on LLT

During a median follow-up of 13.7 years (Q1=11.5 years, Q3=14.9 years), 548 (255 women) out of 3012 participants developed an incident CVD event. The Kaplan-Meier curves for survival free of CVD by LDL groups are displayed in figure 2 (log-rank P<0.0001). Table 2 displays the absolute disease risks by LDL group for the entire sample and stratified by sex and age group (<60 and ≥60 years). Individuals on LLT (group 3 and group 5) had substantial residual cardiovascular risk for incident CVD that is, in the overall sample, approximately three times the absolute disease risk observed in group 1 (LDL below 100 mg/dL, no LLT; table 2 ‘crude event rates’). In individuals below 60 years of age, the absolute disease risks in groups 3 and 5 (on LLT) were almost sixfold and fourfold, respectively, the risk in the referent group (table 2). In the age group ≥60 years, the absolute cardiovascular risk rose for groups on LLT, but also the absolute risk in the referent group 1 increased, so that the relative risks in groups 2–5, compared with group 1, diminished with age. Furthermore, absolute CVD risks were slightly higher in men as compared with women for most LDL groups (table 2).

Request permission

Impact of adjusting for potential covariates, subclinical atherosclerosis of the carotid artery and biomarker levels on residual CVD risk

Compared with reference group 1, LDL groups 3–5 displayed a statistically significantly increased relative hazard for incident CVD in multivariable-adjusted models (figure 3). Adjustment for subclinical atherosclerosis in the carotid arteries only modestly reduced these HRs by up to ~5%, and further adjustment for CRP and BNP had no relevant effect on the effect estimates (table 3).

Request permission

Compared with untreated individuals with similar LDL (groups 1+2 combined), individuals on treatment who reach LDL <130 mg/dL (group 3) had statistically increased hazards for incident CVD in an age- and sex-adjusted model (HR: 1.44, 95% CI 1.06 to 1.96, P=0.02), but this difference was no longer statistically significant on multivariable adjustment (HR: 1.20, 95% CI 0.88 to 1.64, P=0.25). With respect to individuals with LDL ≥130 mg/dL, treated individuals (group 5) were not different from untreated individuals (group 4) with respect to new-onset CVD in age- and sex-adjusted (HR: 1.31, 95% CI 0.92 to 1.87, P=0.13) and in multivariable-adjusted (HR: 1.09, 95% CI 0.76 to 1.55, P=0.65) models.

Principal observations

First, subclinical disease burden and risk for incident CVD rose with increasing LDL category. Second, individuals who reach LDL levels <130 mg/dL on treatment (group 3) as well as individuals with LDL ≥130 mg/dL (groups 4 and 5) displayed greater carotid atherosclerotic disease burden, as compared with untreated individuals with usual LDL <100 mg/dL (group 1). Third, individuals on lipid-lowering medication (groups 5 and 3) had a substantial residual absolute risk for incident CVD that was about three times the risk observed in individuals with LDL <100 mg/dL who were not on treatment. Residual CVD risk rose substantially with age, and men had slightly higher absolute CVD risk than women in most LDL categories. Fourth, part of this increased risk was due to worse cardiovascular risk profiles in groups 3–5 (table 1). However, on adjustment for traditional CVD risk factors, individuals on LLT who reach LDL <130 mg/dL still had about 50% greater hazards for CVD as compared with the referent group. Similarly, groups 4 and 5 displayed statistically significant HRs for CVD (compared with group 1), taking potential confounders into account. Fifth, adjustment for subclinical carotid disease burden only moderately attenuated these HRs (by about 5%) consistent with partial mediation of the residual CVD risk by the burden of subclinical atherosclerosis. Further adjustment for CRP and BNP did not significantly alter the risk estimates.

In the context of the published literature

The association between lipid measures and CVD risk is well established in the epidemiological and clinical literature.17 Numerous clinical trials have demonstrated that lowering of LDL-C through pharmacological or lifestyle interventions reduces the risk for recurrent CVD and mortality in patients with established CVD1 18 or in the primary prevention setting.2 19 20 Furthermore, clinical trials have shown that statins—by lowering LDL—also improve subclinical CVD measures. A recent meta-analysis of 11 clinical trials reported beneficial effects of statins on carotid atherosclerosis.21

Despite clinically and statistically significant reductions in clinical and subclinical CVD risk in the LLT arm compared with the respective comparison groups, many clinical trials reported a relevant residual cardiovascular risk in individuals receiving LLT, with the magnitude of the residual risk depending on the exact patient sample, the duration of follow-up, the type (generation) and dosage of treatment used, the target LDL levels that have been reached, and the exact endpoint investigated in the study.1 18 19 22–25 Clinical trials have relatively strict inclusion and exclusion criteria so that it is unclear whether the observations from clinical trials regarding the residual cardiovascular risk on LLT are applicable to the general population.26

In our unselected sample from the community, we confirm a substantial residual absolute CVD risk in individuals on LLT (group 3 and group 5). Individuals in these groups had about three times the risk of the referent group 1 (untreated LDL <100 mg/dL) in the overall sample. In individuals below 60 years of age, the ratios of the absolute disease risks in those on LLT and those in the referent group 1 were even more pronounced.

Furthermore, we explored the subclinical disease burden at the carotid arteries associated with various LDL levels, stratified by LLT status. Individuals on LLT and with greater LDL levels displayed statistically significantly greater odds for carotid atherosclerosis on ultrasound, including increased IMT and stenoses,13 consistent with other observational studies.27 28 In addition, we evaluated to what extent the increased hazards for CVD in LDL groups 3–5 (figure 3) were attributable to subclinical atherosclerosis at the carotid arteries and to systemic inflammation and neurohormonal activation. Our observations indicate that adjustment for subclinical carotid atherosclerosis only modestly attenuated the relative risk estimates, consistent with the concept that a modest proportion of the increased risk is attributable to subclinical atherosclerosis. Additional adjustment for established biomarkers of systemic inflammation (CRP) and neurohormonal activation (BNP) had no significant effect on the risk estimates.

Important drivers for the elevated subclinical and clinical CVD risk in individuals on LLT with LDL levels <130 mg/dL (group 3) are unfavourable levels for other risk factors, including systolic blood pressure, body mass index, diabetes and HDL in participants in this group (table 1). The greater CVD risk at baseline in individuals on LLT (also reflected in online supplementary figure 1) may reflect indication bias (confounding by indication; higher risk individuals are more likely to be treated by their physicians), as expected in observational (non-randomised) studies assessing residual risk on treatments.29 Indeed, one of the factors contributing to the decision whether LLT is initiated in a given patient is the overall CVD risk over a 10-year time horizon, which takes levels of several risk factors into account.17 On the other hand, and as a clinical consequence, these observations underscore the importance of monitoring all traditional risk factors in individuals on LLT, even if people reach LDL levels below 130 mg/dL.

It also has to be kept in mind that LDL-cholesterol, despite being an established lipid measure and CVD risk factors, might mirror LDL-related residual CVD risk not ideally, particularly in subgroups of the population, where other lipid measures (such as apolipoprotein B (ApoB) concentrations) provide discordant information. In the Women’s Health Study, approximately one in five women were discordant for LDL-C and ApoB; and CVD risk in these individuals was underestimated or overestimated, when estimations were solely based on LDL-C.30 Also in the Framingham Offspring cohort, ApoB levels provided additional information about CVD risk, beyond established lipid measures, in individuals, where the observed ApoB levels were discordant to the expected ApoB concentrations, as predicted based on circulating LDL-C.31 Furthermore, in Jupiter trial participants (individuals with LDL <130 mg/dL and high-sensitivity  C-reactive protein (hsCRP) ≥2.0 mg/L), ApoB levels, but not LDL-C, were associated with incident CVD in the placebo arm.32 In the on-statin arm, however, ApoB failed to reach significance (HR: 1.20, 95% CI 0.98 to 1.46 for incident CVD).32

Strengths and limitations

The strengths of our investigation include the large community-based sample, the careful and comprehensive characterisation of our study participants, including measures of subclinical atherosclerosis at the carotid arteries and of biomarkers of systemic inflammation and neurohormonal activation, and the prospective follow-up for incident CVD (median follow-up, 13.7 years). The following limitations merit consideration. We assigned participants to LDL groups based on a single LDL measurement and a single assessment of the intake of lipid-lowering medication. This might result in some misclassification, which is likely to be non-differential; this would bias us towards the null hypothesis of no difference between the groups in risk of outcome events. In our analyses we did not differentiate between different types of lipid-lowering medication, dosages and durations of treatment. Unfortunately, information on the generation of statin taken or on statin dose was not available in our sample. Furthermore, the prescription of LLT is determined by circulating lipid concentrations and by the overall burden of CVD risk factors. Thus, individuals on LLT in the community might a priori be at higher CVD risk as compared with individuals not receiving such medication (confounding by indication).29 Finally, we have a relatively small number of individuals on statins. However, these individuals were well phenotyped and had a relatively long follow-up period well beyond that of typical clinical trials of LLT.

In conclusion, we observed a substantial residual cardiovascular risk in individuals on lipid-lowering medication in the community, partly explained by an adverse profile of other CVD risk factors and in part (by about 5%) by carotid subclinical atherosclerosis in these people. These observations emphasise the necessity to closely monitor all standard CVD risk factors in individuals on LLT, even when LDL levels are lowered effectively.