Elevated remnant cholesterol and atherosclerotic cardiovascular disease in diabetes: a population-based prospective cohort study

Aims/hypothesis Elevated remnant cholesterol is observationally and causally associated with increased risk of atherosclerotic cardiovascular disease (ASCVD) in the general population. This association is not well studied in individuals with diabetes, who are often included in clinical trials of remnant cholesterol-lowering therapy. We tested the hypothesis that elevated remnant cholesterol is associated with increased risk of ASCVD in individuals with diabetes. We also explored the fraction of excess risk conferred by diabetes which can be explained by elevated remnant cholesterol. Methods We included 4569 white Danish individuals with diabetes (58% statin users) nested within the Copenhagen General Population Study (2003–2015). The ASCVDs peripheral artery disease, myocardial infarction and ischaemic stroke were extracted from national Danish health registries without losses to follow-up. Remnant cholesterol was calculated from a standard lipid profile. Results During up to 15 years of follow-up, 236 individuals were diagnosed with peripheral artery disease, 234 with myocardial infarction, 226 with ischaemic stroke and 498 with any ASCVD. Multivariable adjusted HR (95% CI) per doubling of remnant cholesterol was 1.6 (1.1, 2.3; p=0.01) for peripheral artery disease, 1.8 (1.2, 2.5; p=0.002) for myocardial infarction, 1.5 (1.0, 2.1; p=0.04) for ischaemic stroke, and 1.6 (1.2, 2.0; p=0.0003) for any ASCVD. Excess risk conferred by diabetes was 2.5-fold for peripheral artery disease, 1.6-fold for myocardial infarction, 1.4-fold for ischaemic stroke and 1.6-fold for any ASCVD. Excess risk explained by elevated remnant cholesterol and low-grade inflammation was 14% and 8% for peripheral artery disease, 26% and 16% for myocardial infarction, 34% and 34% for ischaemic stroke, and 24% and 18% for any ASCVD, respectively. LDL-cholesterol did not explain excess risk, as it was not higher in individuals with diabetes. We also explored the fraction of excess risk conferred by diabetes which can be explained by elevated remnant cholesterol. Conclusions/interpretation Elevated remnant cholesterol was associated with increased risk of ASCVD in individuals with diabetes. Remnant cholesterol and low-grade inflammation explained substantial excess risk of ASCVD conferred by diabetes. Whether remnant cholesterol should be used as a treatment target remains to be determined in randomised controlled trials. Graphical Abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-023-06016-0.


Age with left truncation as time scale
Individuals were included from age at baseline until age at event or censoring.Survival age as timescale with exclusion of time before the study is known as age timescale with left truncation, entry age-adjusted age-scale, age timescale with delayed entry, or age timescale with entry at baseline age.This approach adjusts for age at baseline in a non-parametric way (i.e.does not require statistical modelling of the relationship between age and outcome), and as such minimizes risk of bias due to age(1).
To illustrate this, a simulation study by Pencina et.al (in which the entry-age-adjusted age-scale model was considered "the most appropriate" for epidemiological studies) concluded that an incorrect functional form of the age term in time-on-study models may lead to inflated or deflated probabilities of survival and inferior calibration (2).Despite this, they found that regression coefficients for risk factors of interest were very similar in other commonly used models using time-on-study as time scale with parametric adjustment for entry age.As such, age timescale with left truncation and time-on-study timescale with parametric adjustment for age perform equally well for estimation of coefficients, which are used to yield relative risks such as hazard ratios.Cox regressions were additionally adjusted for birth year to adjust for calendar effects (e.g.changes in diagnostics and treatment practices over time).

Sensitivity analysis
In sensitivity analyses, we examined the associations of elevated remnant cholesterol and low-density lipoprotein (LDL) cholesterol with risk of atherosclerotic cardiovascular disease (ASCVD) i) with additional adjustment for body mass index and ASCVD before baseline, ii) with Fine and Gray competing risk regression with death as competing event, iii) using time on study as timescale with adjustment for age using restricted cubic splines, iv) without correction for regression dilution bias, v) additionally adjusting for high-sensitivity C-reactive protein, glucose-lowering medication use, body mass index, apolipoprotein B, high-density lipoprotein (HDL) cholesterol, triglycerides, and LDL triglycerides, vi) using remnant cholesterol and LDL cholesterol calculated from the Martin-Hopkins(3) and the Sampson-NIH(4) formulas, vii) using remnant cholesterol, LDL cholesterol, and LDL triglycerides directly measured using nuclear magnetic resonance spectroscopy, viii) excluding individuals with diabetes type 1, ix) excluding all individuals with ASCVD before baseline, x) stratified by non-fasting plasma glucose, and xi) stratified by sex.

Analysis of explained excess risk
The method used for analysis of explained excess risk is often referred to as mediation analysis.The CMAverse package integrates exposure*mediator interaction in mediation analyses (5).As the association of the mediator with risk of ASCVD may be different in individuals with and without diabetes, interaction is essential to include in analysis of explained excess risk/mediation analysis.This can help estimate the fraction of excess risk of ASCVD in diabetes which may be explained by the explanatory factor/mediator, here termed explained excess risk.Measurement error in remnant cholesterol and C-reactive protein was corrected based on the standard errors from replicate measurements 10 years apart from the Copenhagen City Heart Study (see following section), using quadratic simulation extrapolation with 100 replications (6).

Measurement error correction
Random measurement error biases regression coefficients towards lower estimates (=regression dilution bias).Correction for measurement error was done for remnant cholesterol and LDL cholesterol in all regression analyses using regression calibration (7), except for explained excess risk analyses.Hazard ratios and 95% confidence intervals were corrected for measurement error using repeat measurements from 2,912 individuals without lipid-lowering therapy participating in both the 1991-1994 and 2001-2003 examinations of the Copenhagen City Heart Study (8).Regression calibration, using the same adjustments as in the Cox regressions, was used to yield calibration factors (9).Calibration factors for remnant cholesterol, LDL cholesterol, and high-sensitivity C-reactive protein were 0.48, 0.58, and 0.57.The hazard ratios from the age and sex adjusted model and the multivariable adjusted model, respectively, were corrected by dividing the natural logarithm of the hazard ratio by the corresponding calibration factor.Results from i) original Cox model, ii) additionally adjusted for body mass index, iii) additionally adjusted for apolipoprotein B, iv) additionally adjusted for HDL cholesterol, and v) additionally adjusted for triglycerides.The original Cox model was adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, HDL=High-density lipoprotein, LDL=Low-density lipoprotein, No.=Number.

Hazard ratio (95% CI) per double remnant or LDL cholesterol, additionally adjusted for HDL cholesterol p
Hazard ratio (95% CI) per double remnant or LDL cholesterol, additionally adjusted for triglycerides p 0.9 (0.  Results from i) Cox model using the Friedewald formula (original model), ii) using the Martin-Hopkins formula, and iii) using the  Results from analyses i) excluding individuals with type 1 diabetes, and ii) excluding individuals with ASCVD before baseline.
Cox regression adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, LDL=Low-density lipoprotein, No.=Number.

ESMFig. 6 .
Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study after adjustment for body mass index, apolipoprotein B, HDL cholesterol, and triglycerides.

ESM Fig. 7 .
Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study with different formulas for calculating remnant cholesterol and LDL cholesterol.

ESM Table 1. Baseline characteristics of individuals with diabetes in the Copenhagen General Population Study by statin treatment. Statin use Yes No All
Values are median (inter-quartile range) for continuous variables and number (percentage) for categorical variables.Statin use indicates use of lipid-lowering medications, which was mostly statins.ASCVD=Atherosclerotic cardiovascular disease, comprising peripheral artery disease, myocardial infarction, and ischaemic stroke before baseline, LDL=Low-density lipoprotein ESM

Table 2 . Baseline characteristics of individuals with diabetes in the Copenhagen General Population Study by current smoking.
Values are median (inter-quartile range) for continuous variables and number (percentage) for categorical variables.Statin use indicates use of lipid-lowering medications, which was mostly statins.ASCVD=Atherosclerotic cardiovascular disease, comprising peripheral artery disease, myocardial infarction, and ischaemic stroke before baseline, LDL=Low-density lipoprotein ESM

ESM Figures ESM Fig. 1. Correlation between remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study.
The blue line is linear regression.R 2 is the coefficient of determination in percent.LDL=Low-density lipoprotein.No.=Number

ESM Fig. 2. Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes who were not using statins and who were non-smokers, respectively, from the Copenhagen General Population Study Cox
regression multivariable adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status (except for analyses in non-smokers), cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, LDL=Low-density lipoprotein, No.=Number.

. Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes who were using statins and who were smokers, respectively, from the Copenhagen General Population Study.
Cox regression adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status (except for analyses in smokers), cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, LDL=Low-density lipoprotein, No.=Number.

. Sensitivity analyses of risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study.
Results from i) original multivariable adjusted Cox model, ii) additionally adjusted for BMI and ASCVD before baseline, iii) Fine and Gray competing risk regression with death as competing event, iv) using time on study as timescale, and v) without correction for regression dilution bias.The original Cox model was adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, BMI=Body mass index, CI=Confidence interval LDL=Low-density lipoprotein, No.=Number.EndpointNo.

Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study after adjustment for high-sensitivity C-reactive protein and antidiabetic medication use
Results from i) original Cox model, ii) additionally adjusted for high-sensitivity C-reactive protein, iii) additionally adjusted for antidiabetic medication use.The original Cox model was adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).Antidiabetic medication use was either no antidiabetic medication, only insulin, only other antidiabetic medication, or both insulin and other antidiabetic medication.ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, HDL=High-density lipoprotein, LDL=Low-density lipoprotein, No.=Number.

95% CI) per double remnant or LDL cholesterol, multivariable adjusted p Hazard ratio (95% CI) per double remnant or LDL cholesterol, additionally adjusted for body mass index p
Sampson-NIH formula.Adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval LDL=Low-density lipoprotein, No.=Number.

Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study after adjustment for LDL triglycerides.
The Cox model was adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, LDL triglycerides, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, HDL=Highdensity lipoprotein, LDL=Low-density lipoprotein, No.=Number.

Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study by groups of non-fasting plasma glucose.
Results from i) individuals with non-fasting plasma glucose <6.5 mmol/L (<median), and ii) individuals with non-fasting plasma glucose ≥6.5 mmol/L (≥median).Cox regression adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, LDL=Low-density lipoprotein, No.=Number.

11. Risk of ASCVD per doubling of directly measured remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study.
Cox regression adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting glucose, and directly measured LDL cholesterol (in remnant cholesterol analyses) or directly measured remnant cholesterol (in LDL cholesterol analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence Interval, LDL=Lowdensity lipoprotein, No.=Number.

12. Risk of ASCVD per doubling of directly measured remnant cholesterol and LDL triglycerides in individuals with diabetes from the Copenhagen General Population Study.
Cox regression adjusted for age, sex, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, birth year, non-fasting plasma glucose, directly measured LDL cholesterol, LDL triglycerides (in remnant cholesterol analyses) or directly measured remnant cholesterol (in LDL triglycerides analyses).ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence Interval, LDL=Low-density lipoprotein, No.=Number.

Risk of ASCVD per doubling of remnant cholesterol and LDL cholesterol in individuals with diabetes from the Copenhagen General Population Study in men and women
Results for i) men only, and ii) women only, from Cox regression multivariable adjusted for age, systolic blood pressure, diastolic blood pressure, smoking status, cumulative smoking, non-fasting plasma glucose, birth year, and LDL cholesterol (in remnant cholesterol analyses) or remnant cholesterol (in LDL cholesterol analyses).Number of events per 1,000 person years is from unadjusted Poisson regression.ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, LDL=Low-density lipoprotein, No.=Number.

14. Excess risk of ASCVD for diabetes in individuals who were using statins and who were smokers, respectively, from the Copenhagen General Population Study.
Cox regression adjusted for age, sex, smoking status, cumulative smoking, and birth year.ASCVD=Atherosclerotic cardiovascular disease, CI=Confidence interval, LDL=Low-density lipoprotein, No.=Number.

18. Violin plots showing densities of remnant cholesterol, LDL cholesterol, and high-sensitivity C-reactive protein in individuals without diabetes, individuals with diabetes and non-fasting plasma glucose <6.5 mmol/L, and individuals with diabetes and non-fasting plasma glucose ≥6.5 mmol/L, respectively, in the Copenhagen General Population Study.
CRP= C-reactive protein, IQR=Interquartile range, LDL=Low-density lipoprotein, No.=Number.