Further methodological considerations in the assessment of risk-weighted apolipoprotein B

This commentary refers to ‘Risk-weighted apolipoprotein B: a novel summary metric outperforming traditional lipid biomarkers in predicting coronary heart disease’, by M. B. Rehman et al., https://doi.org/10.1093/eurheartj/ehaf1124 and the discussion piece ‘Risk-weighted apoB: the evidence on the scales’, by M. B. Rehman and E. Björnson, https://doi.org/10.1093/eurheartj/ehag260. We read with great interest the article by Rehman et al. introducing risk-weighted apolipoprotein B (RW-apoB) as a novel measure integrating the differential atherogenicity of LDLs, triglyceride-rich lipoproteins, and lipoprotein(a) Lp(a) to improve coronary heart disease (CHD) risk prediction.1 The study is well-justified biologically and strengthened by validation across multiple large population-based cohorts. The authors appropriately acknowledge several important limitations, including the use of non-fasting triglyceride measurements in the UK Biobank, truncation of Lp(a) values due to assay constraints, and the need for further evaluation of RW-apoB in statin-treated individuals and across diverse ethnic groups. After careful review, we respectfully suggest that several additional methodological aspects warrant consideration, as they may influence interpretation of the findings and their potential clinical application. First, no formal power calculation has been reported. Although overall sample sizes are large, the absence of a priori or post hoc power considerations remains relevant, particularly for subgroup, interaction, and treatment-stratified analyses, where effect estimates may be more sensitive to limited statistical power. Second, RW-apoB is a composite variable mathematically derived from apoB, triglycerides, and Lp(a). While this construction is conceptually appealing, the potential for collinearity or redundancy within multivariable Cox models has not been discussed. When mathematical linkage between component predictors is not adequately addressed, derived estimators may appear to outperform their individual components, reflecting statistical dependency rather than true biological integration.2 Formal assessment of collinearity (e.g. variance inflation factors) and sensitivity analyses comparing RW-apoB with models including its individual components would help clarify whether improvements in discrimination reflect biological integration rather than statistical dependency. Sensitivity analyses restricted to hard CHD endpoints (MI and CHD death) could further clarify robustness. Third, the fundamental assumptions underlying the Cox proportional hazards models have not been explicitly reported. Documenting proportional hazards tests, assessing linearity for continuous predictors, and evaluation of interaction terms—particularly with lipid-lowering therapy—would strengthen confidence in the robustness of the reported associations.3 Fourth, definitions of CHD outcomes vary across cohorts, encompassing various combinations of myocardial infarction, angina, coronary revascularization, and CHD-related death. While this reflects real-world registry data, heterogeneity in outcome definitions can influence effect estimates and inter-cohort comparability. Harmonized outcome analyses or competing-risk approaches can help assess the impact of this variability.