A Latent Autoantibody Axis Associated with Vascular Vulnerability in Ischemic Stroke: Integrated Statistical and Machine-Learning Analysis

TL;DR

This study explores how autoantibodies may reflect vascular risk in stroke patients, finding a potential biological signal that could help assess stroke vulnerability.

Contribution

The study identifies a latent autoantibody axis linked to vascular vulnerability in ischemic stroke, using integrated statistical and machine-learning methods.

Findings

Antibody markers did not improve overall model performance but were influential in machine-learning interpretations.

PCA identified a dominant antibody component strongly associated with stroke risk after age adjustment.

Unsupervised clustering revealed a high-risk subgroup with elevated antibody levels.

Abstract

Ischemic stroke remains a major cause of mortality and long-term disability worldwide, and improved strategies for identifying individuals at elevated vascular risk are needed. Serum autoantibodies have emerged as potential biomarkers reflecting vascular injury and immune activation; however, their integrative biological significance and incremental predictive value beyond established clinical risk factors remain unclear. We analyzed 833 participants, including patients with acute ischemic stroke (AIS) or transient ischemic attack (TIA) and healthy controls. Serum levels of anti-PDCD11 antibody (Ab), anti-DNAJC2 antibody, and anti-PAI-1 (SERPINE1) antibody were quantified, and multivariable logistic regression and machine-learning (ML) models (logistic regression and random forest) were constructed using clinical variables with and without antibody markers. Model performance was evaluated using cross-validation, bootstrap-derived confidence intervals, calibration metrics, and reclassification indices. Model interpretability analyses, principal component analysis (PCA), unsupervised clustering, and propensity score matching were performed to explore latent biological structures. Clinical-only models demonstrated excellent discrimination (bootstrap Area Under the Curve (AUC) 0.917 for random forest and 0.919 for logistic regression). The addition of antibody markers yielded similar performance (AUC 0.913 and 0.923, respectively) without evidence of meaningful improvement in reclassification. However, SHapley Additive exPlanations (SHAP) analysis identified antibody markers as influential contributors following major clinical risk factors. PCA revealed a dominant antibody component explaining approximately 79% of the variance, which remained independently associated with stroke after age adjustment. Unsupervised clustering further identified a high-risk subgroup characterized by consistently elevated antibody levels. These findings support the presence of a latent antibody axis associated with vascular vulnerability. Although antibody markers did not substantially enhance global predictive performance, they captured integrated biological signals reflecting cumulative vascular and immunological stress. Autoantibody profiling may complement conventional risk assessment by improving biological characterization of stroke susceptibility. Prospective validation in independent cohorts is required prior to clinical implementation.

Figures