IgG4-related disease (IgG4-RD) is a chronic immune-mediated disease with heterogeneity. In this study, we used machine-learning approaches to characterize the immune cell profiles and to identify the heterogeneity of IgG4-RD. The XGBoost model discriminated IgG4-RD from HCs with an area under the receiver operating characteristic curve of 0.963 in the testing set. There were two clusters of IgG4-RD by k-means clustering of immunological profiles. Cluster 1 featured higher proportions of memory CD4⁺T cell and were at higher risk of unfavorable prognosis in the follow-up, while cluster 2 featured higher proportions of naïve CD4⁺T cell. In the multivariate logistic regression, cluster 2 was shown to be a protective factor (OR 0.30, 95% CI 0.10–0.91, P = 0.011). Therefore, peripheral immunophenotyping might potentially stratify patients with IgG4-RD and predict those patients with a higher risk of relapse at early time.

IgG4相关疾病（IgG4-RD）是一种具有异质性的慢性免疫介导疾病。在这项研究中，我们使用机器学习方法来表征免疫细胞特征并识别 IgG4-RD 的异质性。XGBoost 模型将 IgG4-RD 与 HC 区分开来，测试集中的受试者工作特征曲线下面积为 0.963。通过免疫学特征的 k 均值聚类，有两个 IgG4-RD 簇。簇 1 具有较高比例的记忆 CD4 ⁺ T 细胞，并且在随访中出现不良预后的风险较高，而簇 2 具有较高比例的初始 CD4 ⁺ T 细胞。在多变量逻辑回归中，聚类 2 显示为保护因素（OR 0.30，95% CI 0.10–0.91，P = 0.011）。因此，外周免疫表型分析可能会对 IgG4-RD 患者进行分层，并在早期预测那些复发风险较高的患者。