Unmasking residual cardiovascular risk: the paradoxical interaction between remnant cholesterol and calculated LDL-C in a tertiary-care cohort

BACKGROUND: Residual atherosclerotic cardiovascular disease (ASCVD) risk often remains even after low-density lipoprotein cholesterol (LDL-C) levels have been brought down to target levels. Remnant cholesterol (RC) and inflammation have been increasingly linked to the residual risk. We aimed to investigating whether the discriminative value of RC the ability of RC to discriminate and its claimed interactions with LDL-C are due to a real clinical phenotype or are affected by formula-dependent biases between the Friedewald and Sampson-NIH equations. METHODS: We performed a cross-sectional analysis of consecutively tested adults (n = 3,342) using residual serum samples from routine clinical monitoring. To reduce analytical variability, all lipid profiles were analyzed using a single, dedicated reagent lot. We contrasted risk models with Friedewald-calculated versus Sampson-NIH-calculated LDL-C to assess equation-dependent differences. Lipid parameters, hemoglobin A1c (HbA1c), estimated glomerular filtration rate (eGFR), and C-reactive protein (CRP) were measured. ASCVD was defined using International Classification of Diseases, 10th Edition (ICD-10) codes. Missing covariate data were handled using multiple imputation by chained equations (m = 50), with additional complete-case sensitivity analyses for CRP-related models. To reduce bias, the observed ASCVD status was included as an auxiliary variable; the outcome itself was not imputed. The discriminative performance of nested logistic regression models was assessed through the pooled area under the receiver operating characteristic curve (AUC) and pooled DeLong p-values. RESULTS: The primary clinical focus was the presence of documented pre-existing ASCVD diagnoses, identified in 11.4% of the cohort, while 9.4% of participants met the criteria for atherogenic dyslipidemia (AD). In the primary analysis with Friedewald LDL-C, we detected a statistically significant (p < 0.001) negative interaction between LDL-C and RC, while logCRP remained an independent correlate in the adjusted model. Interestingly, when we verified this using the more accurate Sampson-NIH equation to minimize the possibility that the result would be solely due to calculation bias, the paradoxical interaction was still statistically significant (p = 0.003) along with a strong model performance (AUC: 0.729). This indicates that the interaction is not entirely explained by the mathematical artifact of the Friedewald formula, but rather represents a consistent statistical pattern in this cohort. CONCLUSION: RC adds statistically significant value to risk discrimination. The continuous inverse relationship of LDL-C with high RC identifies a statistical pattern consistent with persistent atherogenic burden despite apparently optimal calculated LDL-C levels. Awareness of this potential suppressor effect may aid in refining risk stratification in tertiary-care settings.