Work

Conventional lipid augmentation in simplified cardio-renal-metabolic clustering after percutaneous coronary intervention: a two-cohort exploratory analysis

Year: 2026
Type: article
Abstract: Background Although lipid biomarkers are central to cardiovascular risk stratification, whether they provide incremental information for unsupervised cardio–renal–metabolic (CRM) cluster separation after percutaneous coronary intervention (PCI) remains uncertain. Methods K-means clustering was applied to the PCI TOTAL registry ( n = 1,240) using four CRM variables: age, left ventricular ejection fraction, fasting glucose and serum creatinine; K = 3 was retained based on parsimony, a larger minimum cluster size than K = 5, interpretability of the clustering-variable profiles, and bootstrap resampling stability. External evaluation used the MASS-SYNTAX PCI subgroup ( n = 532). Lipid augmentation was assessed against a correlation-preserving permutation reference and a dimension-balanced lipid-domain representation, and stability was quantified by bootstrap resampling on a common evaluation set. Mortality was analyzed as a binary outcome because harmonized fixed-time endpoints were unavailable. Results Three clusters were identified: A (Younger/Preserved-Function, 52.1%; mortality 2.6%), B (Metabolic-Dominant, 11.9%; 10.8%) and C (CRM-Renal, 36.0%; 12.3%). Adding TC, LDL-C and HDL-C reduced the silhouette score from 0.221 to 0.156 and reorganized assignments (adjusted Rand index 0.176); a dimension-balanced lipid-domain representation attenuated but did not reverse this reduction. The observed seven-variable silhouette score (0.156) was lower than the correlation-preserving permutation reference (mean 0.184 ± 0.006). The overall partition showed stability under bootstrap resampling (mean Jaccard 0.862–0.902; mean adjusted Rand index 0.809); the youngest cluster retained the lowest mortality in 997/1,000 resamples, whereas a strictly increasing age-ordered gradient was preserved in 655/1,000. In the external cohort, independent re-clustering and strict centroid transfer showed limited individual-level agreement (adjusted Rand index 0.224; label-matched concordance 64.3%). In exploratory univariable analyses, no conventional lipid measure was significantly associated with observed all-cause mortality. Conclusions In these two historical PCI cohorts, adding conventional lipid measures did not improve silhouette-based separation within the evaluated four-variable CRM clustering model. The resulting clusters showed internal stability but limited individual-level agreement between external evaluation strategies. These findings are exploratory and specific to the clustering approaches examined. They do not diminish the established clinical importance of lipid biomarkers or lipid-lowering therapy.
Language: English
Field: Medicine
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7235301795
FWCI:
Cites: 32
Cited by:
Related to: 10
Parts: 1