Incremental value of renal, metabolic, and inflammatory markers for prevalent self-reported cardiovascular disease across NHANES cycles: A repeated cross-sectional validation study

Renal dysfunction, metabolic stress, and systemic inflammation are associated with cardiovascular disease (CVD), but their incremental value beyond routine clinical factors is uncertain. This repeated cross-sectional study used 7 National Health and Nutrition Examination Survey (NHANES) cycles (2005-2018) for development and leave-one-cycle-out validation. The primary outcome was prevalent self-reported CVD. A locked clinical core was extended with estimated glomerular filtration rate plus log2 urine albumin-to-creatinine ratio (renal), triglyceride-glucose waist circumference index (TyG-WC; metabolic), systemic inflammation response index (inflammatory), or all markers. Elastic Net was primary and XGBoost secondary. Preprocessing, imputation, and full-grid tuning were confined to training cycles. Survey-aware 95% confidence intervals (CIs) used 2000 Rao-Wu bootstrap replicates. NHANES 2021-2023 underwent a separate temporal assessment, corrected after inspection of initial outputs, using development-trained pipelines without temporal tuning or recalibration. Development examination and fasting samples included 37,482 adults (4222 CVD events) and 15,956 adults (1813 events), respectively. The Elastic Net core area under the receiver operating characteristic curve (AUROC) was 0.8490 (95% CI, 0.8410-0.8560). Renal markers improved AUROC by 0.0046 (0.0031-0.0062) and area under the precision-recall curve (PR-AUC) by 0.0161 (0.0101-0.0223); the combined extension improved AUROC by 0.0072 (0.0039-0.0106). In the corrected 2021-2023 assessment, phlebotomy and fasting samples included 5755 adults (739 events) and 2940 adults (355 events). Core AUROC was 0.8533 (0.8313-0.8753). Renal incremental AUROC was 0.0041 (0.0001-0.0081) and PR-AUC was 0.0217 (0.0093-0.0310); combined incremental AUROC was 0.0074 (0.0016-0.0124). TyG-WC alone showed no clear increment. XGBoost gains were less consistent across cycles and temporal metrics. Renal and combined extensions showed modest, algorithm-dependent gains in classifying prevalent self-reported CVD. The corrected temporal assessment provided supportive evidence for Elastic Net. These findings describe cross-sectional association patterns; prospective studies with adjudicated outcomes and defined clinical decisions are needed to evaluate clinical utility.