Data Sheet 1_An oral microbiome model for predicting atherosclerotic cardiovascular disease.docx

ObjectiveThis study aimed to construct a predictive model for the early onset of atherosclerotic cardiovascular disease (ASCVD) by integrating oral microbiome data with traditional clinical risk factors.
MethodsA retrospective study was conducted involving participants aged 50–70 years without pre-existing ASCVD. The patients were divided into a training set and a validation set at a ratio of 7:3 by the complete randomization method. The characteristics of the oral microbiome were characterized by 16S rRNA/metagenomic sequencing. In the training set, univariate analysis and multivariate Logistic regression analysis were applied to screen predictive variables, and Random Forest (RF), Gradient Boosting (GB), and K-nearest Neighbor (KNN) were constructed. The receiver operating characteristic (ROC) curve was validated. The model performance was evaluated by net reclassification improvement (NRI) and integrated discrimination improvement (IDI).
ResultsA total of 331 patients were enrolled and randomly divided into a training set (n=231) and a validation set (n=100). 40 out of 331 participants experienced major adverse cardiovascular events (MACE). Multivariate Logistic regression analysis confirmed that age, relative abundance of Fusobacterium nucleatum, Prevotella, Porphyromonas, Leptotrichia, Streptococcus and Actinomyces were significantly associated with ASCVD event risk (all P < 0.05). Three machine learning models (RF, GB, and KNN) were constructed, with the RF model achieving the highest predictive performance. The AUC values of the RF, GB, and KNN models in the training set were 0.888 (95% CI: 0.818-0.958), 0.823 (95% CI: 0.745-0.901), and 0.812 (95% CI: 0.727-0.898) respectively, and in the validation set were 0.845 (95% CI: 0.740-0.951), 0.746 (95% CI: 0.621-0.871), and 0.767 (95% CI: 0.647-0.887) respectively. Additionally, the integrated model showed significant improvements in net reclassification improvement (NRI = 0.315, P < 0.05) and integrated discrimination improvement (IDI = 0.227, P < 0.05) compared to traditional clinical models.
ConclusionThe integration of the oral microbiome and clinical data can improve the accuracy of the ASCVD risk prediction model, providing a novel biomarker strategy for primary cardiovascular prevention.

研究目的：本研究旨在整合口腔微生物组数据与传统临床危险因素，构建动脉粥样硬化性心血管疾病（atherosclerotic cardiovascular disease，ASCVD）早发预测模型。
研究方法：本研究为回顾性研究，纳入年龄50~70岁、无既往ASCVD病史的受试者。采用完全随机化方法按7:3的比例将受试者划分为训练集与验证集。通过16S rRNA/宏基因组测序（16S rRNA/metagenomic sequencing）对口腔微生物组特征进行表征。在训练集中，采用单因素分析与多因素Logistic回归分析筛选预测变量，并构建随机森林（Random Forest，RF）、梯度提升（Gradient Boosting，GB）及K近邻（K-nearest Neighbor，KNN）模型。对受试者工作特征（receiver operating characteristic，ROC）曲线进行验证，采用净重新分类改善（net reclassification improvement，NRI）与综合判别改善（integrated discrimination improvement，IDI）评估模型性能。
研究结果：本研究共纳入331例受试者，经完全随机分组后分为训练集231例、验证集100例。331例受试者中共有40例发生主要不良心血管事件（major adverse cardiovascular events，MACE）。多因素Logistic回归分析证实，年龄、具核梭杆菌（Fusobacterium nucleatum）、普雷沃菌属（Prevotella）、卟啉单胞菌属（Porphyromonas）、纤细杆菌属（Leptotrichia）、链球菌属（Streptococcus）及放线菌属（Actinomyces）的相对丰度与ASCVD事件风险显著相关（所有P<0.05）。本研究构建了三种机器学习模型（RF、GB及KNN），其中RF模型的预测性能最优。训练集中RF、GB、KNN模型的AUC值分别为0.888（95%CI：0.818~0.958）、0.823（95%CI：0.745~0.901）及0.812（95%CI：0.727~0.898）；验证集中分别为0.845（95%CI：0.740~0.951）、0.746（95%CI：0.621~0.871）及0.767（95%CI：0.647~0.887）。此外，相较于传统临床模型，整合模型的净重新分类改善（NRI=0.315，P<0.05）与综合判别改善（IDI=0.227，P<0.05）均具有显著提升。
结论：口腔微生物组与临床数据的整合可提升ASCVD风险预测模型的准确性，为心血管疾病一级预防提供了全新的生物标志物策略。