Development and external validation of a residual atherogenic lipid burden score for incident stroke: prospective analyses in chinese and english ageing cohorts

Routine lipid fractions incompletely summarize atherogenic cholesterol and triglyceride-rich dyslipidemia. We developed a Residual Atherogenic Lipid Burden Score (RALBS) integrating complementary information from non-HDL-C, the atherogenic index of plasma (AIP), and HDL-C, and assessed its association with incident stroke. In CHARLS, adults aged at least 45 years without stroke were followed from 2011 to 2020. RALBS combined standardized non-HDL-C, atherogenic index of plasma (AIP), and HDL-C using direction-constrained elastic-net Cox regression. Internal validation used 500 bootstrap samples. Baseline and 2015 landmark cumulative associations were assessed using Cox and competing-risk models. Incremental prediction compared RALBS with simpler lipid markers against a common clinical-plus-LDL-C reference. For external evaluation, the unchanged CHARLS score and prediction coefficients were applied to stroke-free ELSA wave-6 participants and followed through wave 10; five-year discrimination and calibration of absolute-risk estimates were evaluated. The CHARLS exposure cohort included 10,505 participants and 938 incident strokes; the primary complete-case model included 8,670 participants and 796 events. Per development-cohort SD, RALBS was associated with stroke after clinical adjustment (HR 1.15, 95% CI 1.07–1.23), as was cumulative RALBS after the 2015 landmark (HR 1.15, 95% CI 1.04–1.26). Explicit HDL-C was selected in 53.0% of bootstrap samples; the two-component ablation score correlated strongly with RALBS ( r = 0.997). Adding RALBS or AIP to clinical factors and LDL-C yielded similar CHARLS five-year AUCs (0.7404 and 0.7405). In ELSA, the wave-6-to-10 clinical model included 4,537 participants and 192 strokes; the HR per frozen CHARLS SD was 1.26 (1.07–1.48) ( P = 0.006). Among 4,488 participants eligible for prediction transport, adding RALBS increased the five-year AUC from 0.643 to 0.654; the observed-to-expected risk ratio was 0.834 and the calibration slope was 0.561. In ELSA, positive five-year NRI and IDI point estimates for RALBS were accompanied by wide confidence intervals. Higher baseline and sustained RALBS were associated with incident stroke in CHARLS, and the frozen score showed a directionally consistent association in ELSA. External evaluation showed an increase in discrimination after adding RALBS, while the absolute-risk estimates indicated a need for recalibration before clinical application. In CHARLS, the incremental predictive contribution of RALBS was comparable to that of AIP. Not applicable.