1570: Memory CD4⁺T Cell Signature Is Associated with Unfavorable Prognosis in IgG4-related Disease: Risk Stratification by Machine-learning

Yuxue Nie¹, zheng liu¹, Wei Cao², Taisheng Li² and Wen Zhang³, ¹Department of Rheumatology, Peking Union Medical College Hospital,Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China, ²Department of Infectious Diseases, Peking Union Medical College Hospital,Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China, ³Peking Union Medical College Hospital, Beijing, China

Background/Purpose: IgG4-related disease (IgG4-RD) is a chronic immune-mediated disease with significant heterogeneity. A better understanding of the variance in the immune signature of IgG4-RD will contribute to the treatment of IgG4-RD. In this study, we used machine-learning approaches to demostrate the immune cell profiles and identify the heterogeneity of IgG4-RD.

Methods: Demographic and clinical data of IgG4-RD patients (2019 ACR/EULAR classification criteria) including organ involvement, disease activity (Responder Index) and laboratory tests, were collected respectively. The immunophenotype of PBMC of IgG4-RD patients and healthy controls (HCs) was determined by flow cytometry and univariate analysis to compare percentages of immune cell subsets between IgG4-RD and HCs. EXtreme Gradient Boosting (XGBoost) was applied for the construction of the classification model between IgG4-RD and HCs. To visualize the importance of the features in the XGBoost model, SHapley Additive exPlanations (SHAP) value in the model was shown. K-means clustering with the proportions of T cell subsets and NK cells to identify the heterogeneity of IgG4-RD. Kaplan-Meier analysis and logistic regression were used to investigate the prognosis and potential risk factors for the likelihood of relapse in IgG4-RD. This study was approved by the Ethics Committee of Peking Union Medical College Hospital (PUMCH).

Results: This study included 148 patients with IgG4-RD (median [IQR] age 56.0 [46.8, 63.0] years, 57.4% male) and 188 sex/age-matched HCs. The median follow-up time of those patients was 12 months (IQR [6–12] months). The XGBoost model discriminated patients with IgG4-RD from HCs with an area under the receiver operating characteristic curve (AUC) of 0.963 in the testing set. The top three features identified by SHAP value in the XGBoost model were naïve CD4⁺T, DR⁺CD8⁺T, and CD4⁺T cells. By K-means clustering, IgG4-RD patients were classified into two clusters. Cluster 1 (N = 99) featured a higher proportion of memory CD4⁺T cells, and cluster 2 (N = 49) featured a higher proportion of naïve CD4⁺T cells accordingly. The two clusters shared similar clinical characteristics and the initial or maintenance treatment. The unfavorable prognosis was defined as relapse of IgG4-RD or the decision of enhancing treatment intensity (increased dose of glucocorticoids or switch to more powerful immunosuppressors) by experts in rheumatology during the follow-up visits. In the Kaplan-Meier analysis, cluster 1 was more likely to have an unfavorable prognosis in the follow-up. Furthermore, in the multivariate logistic regression, cluster 2 was shown to be a protective factor.

Conclusion: Immune cell signature with machine-learning approaches was efficient in the discrimination of IgG4-RD patients and HCs. Immunophenotyping of IgG4-RD may help in the identification of patients with the potential to relapse and attention should be paid to those patients with elevated proportions of memory CD4⁺T cells.

Disclosures: Y. Nie, None; z. liu, None; W. Cao, None; T. Li, None; W. Zhang, None.