Lipoproteins predicting coronary lesion complexity in premature coronary artery disease: a supervised machine learning approach.

Abstract

Introduction: We aimed to assess the usefulness of lipoprotein(a) [Lp(a)] and LDL-C levels as potential predictors of coronary lesions' complexity in patients with premature coronary artery disease (pCAD).

Methods: This study enrolled 162 consecutive patients with pCAD undergoing coronary angiography. The SYNTAX score (SS) was used to assess coronary lesions' complexity. Linear discriminant analysis (LDA) was employed to construct a multivariate classification model enabling the prediction of coronary lesions' complexity in SS.

Results: The Lp(a) levels among patients with SS ≥ 23 and with SS 1-22 were significantly higher than those with SS = 0 (p = 0.021 and p = 0.027, respectively). The cut-off point for the Lp(a) level of 63.5 mg/dl discriminated subjects with SS ≥ 23 from those with SS ≤ 22 (sensitivity 0.546, specificity 0.780; AUC 0.620; p = 0.027). An LDA-based model involving the Lp(a) level, age, sex and LDL-C provided improved discrimination performance (sensitivity 0.727, specificity 0.733, AUC 0.800; p = 0.0001).

Conclusions: Lp(a) levels in pCAD patients are associated with the advancement of coronary artery lesions in SS patients. An Lp(a) level of 63.5 mg/dl can be the cut-off point for the identification of subjects with SS ≥ 23. LDA-based modelling using Lp(a), LDL-C, age and gender may be an applicable tool for the preliminary identification of patients at risk of more complex coronary artery lesions.