A machine learning–derived Lp(a) threshold far below current guideline levels predicts cardiovascular events in japanese PCI patients

Abstract

According to the ESC/EAS guidelines, elevated lipoprotein(a) [Lp(a)] levels (>105 nmol/L) are considered a risk factor for cardiovascular events. However, Lp(a) levels vary across ethnicities and are lower in East Asian populations, for whom an appropriate risk threshold has not been established.

This study aimed to evaluate the prognostic significance of Lp(a) and to identify a population-specific Lp(a) cutoff value in Japanese patients undergoing percutaneous coronary intervention (PCI), using the CLIDAS-PCI database.

The CLIDAS-PCI database is a multicenter registry in Japan. Among 9,690 registry patients, we analyzed 474 patients who underwent PCI between April 2014 and March 2019 and had Lp(a) levels measured prior to PCI. A conditional inference tree–based machine learning approach was applied to estimate an optimal Lp(a) cutoff, and patients were stratified into low and high Lp(a) groups according to the derived threshold. The primary endpoint was major adverse cardiac and cerebrovascular events plus (MACCE+), defined as a composite of cardiovascular death, non-fatal myocardial infarction, stroke, and coronary revascularization.

In the overall cohort, the median age was 70 years, 75% were male, and the median Lp(a) concentration was 64.7 [30.1–128.5] nmol/L. During a median follow-up of 2.5 years, MACCE+ occurred in 26%. Conditional inference tree analysis identified a data-driven Lp(a) cutoff of 22.5 nmol/L. Based on this cutoff, patients were stratified into a low Lp(a) group (n = 97) and a high Lp(a) group (n = 377). Baseline clinical characteristics were largely comparable between the two groups stratified by this cutoff; however, the high Lp(a) group exhibited lower estimated glomerular filtration rates (eGFR) and more frequent statin use. Kaplan–Meier analysis demonstrated that patients in the high Lp(a) group experienced significantly higher MACCE+ event rates compared with those in the low Lp(a) group (log-rank p = 0.047; Figure 1). Importantly, in a multivariable Cox proportional hazards model adjusted for age, sex, body mass index, blood pressure, eGFR, hemoglobin A1c, LDL cholesterol, and statin use, elevated Lp(a) remained independently associated with an increased risk of MACCE+ (hazard ratio, 2.00; 95% confidence interval, 1.08–3.72). Furthermore, a significant interaction was observed between Lp(a) category and renal function status (p for interaction = 0.02), indicating that the prognostic impact of elevated Lp(a) was more pronounced among patients with preserved renal function (eGFR >60 mL/min/1.73 m²).

A machine learning–derived cutoff of 22.5 nmol/L stratified cardiovascular risk, well below current guideline thresholds, particularly in patients with preserved renal function. These findings suggest that ethnicity-specific Lp(a) thresholds may improve risk identification in Japanese populations.