ESC Congress 2026 Abstract · Oct 26, 2026
Understanding discrepancies in trial-level surrogacy of LDL-cholesterol reduction for cardiovascular risk: a methodological appraisal and updated meta-regression
By Chen XG, Geusens B, Dobrynina M et al.
Background/Introduction: LDL-cholesterol (LDL-C) is a strong prognostic biomarker for cardiovascular (CV) disease, supported by clinical trials, CTT meta-analyses, and Mendelian randomisation studies. Whether treatment effects on LDL-C predict CV risk reduction at trial level remains contested: Silverman et al. and others reported strong associations, while Byrne et al. and Liaigre et al. found weak surrogacy. Though Liaigre et al. discussed discrepancies and their approach aligns with the HTA surrogate validation framework, how the targeted estimand, limitations of trial-level data, and specific modelling choices interact to produce divergent conclusions warrants further examination. Purpose: We appraised the estimand, data and analytical factors driving conflicting surrogacy conclusions. An systematic literature review (SLR) across lipid-lowering therapy (LLT) classes served to illustrate these factors and contributed to updated results. Methods: We contrasted key published meta analyses on included trials, LDL-C and CV outcome and treatment effect definition, surrogacy estimand and its analytical methods — effect measure, intercept and other modelling specification, and metrics used to characterize the relationship of LDL-C and CV risk reduction. We extended the Silverman SLR, with post-2016 RCTs (including non-statin trials) and performed mixed-effects meta-regression with and without intercept. Results: Discrepancies arise from interacting factors. SLR design, outcome and treatment effect definitions create materially different datasets, even in overlapping trials. Modeling choices further compound differences. Within specific classes, drugs that achieve comparable LDL-C reductions may render linear meta-regression inappropriate for surrogacy analyses. R2 in the meta-analytic context is not a direct measure of prediction nor correlation, may not address the estimand of interest, and is sensitive to trial heterogeneity and range restriction. We discuss a few complementary metrics. Our expanded SLR (61 RCTs; n=276,391) found each 1 mmol/L LDL-C reduction associated with a 22% relative risk (RR) reduction in major vascular events modelling without intercept (consistent with Silverman and CTT assumptions) and 15% RR with intercept. Discussion/Conclusion: Estimating a biological gradient per mmol/L across therapies and predicting benefit for a future trial are distinct causal questions – as is modelling similar drugs across trial designs vs. different drugs across similar trials - yet typical surrogacy analyses collapse them. We found that conflicting surrogacy conclusions are driven primarily by the estimand targeted, compounded by the analytical and data factors described above — not by absence of association. Trial-level analysis has limited ability to confirm or refute surrogacy, but provides evidence of reference value on the magnitude of the LDL-C and CV association — an association our expanded analysis reaffirms.