Background To evaluate the comparative value of traditional clinical features and novel immune-inflammatory indices in assessing disease activity in primary Sjögren's Disease (SjD), and to construct a robust classification model for moderate-to-severe disease activity using a machine learning approach. Methods This cross-sectional study included 18,738 SjD patients from the Chinese Rheumatism Data Center (CRDC) multicenter prospective cohort. Spearman rank correlation and local polynomial regression (LOESS) were used to explore the relationship between systemic inflammatory indices and the EULAR Sjögren's syndrome disease activity index (ESSDAI). To identify moderate-to-severe disease activity (ESSDAI ≥ 5), data were randomly split into training and validation sets (7:3). Classification models were developed using various machine learning techniques. Results Higher ESSDAI scores were significantly associated with elevated levels of both traditional markers (ESR, CRP, IgG) and novel composite indices (NLR, CAR, CLR, IBI) (all p < 0.001). However, these indices exhibited only weak-to-moderate linear correlations with ESSDAI and displayed non-linear saturation trends. In multivariable logistic regression, traditional markers remained independent risk factors, whereas all novel composite indices lost statistical significance, yielding a poor validation area under the curve (AUC) of 0.678. Conversely, the XGBoost model demonstrated modestly favorable overall performance Feature importance analysis revealed that patient-reported outcomes, hemoglobin, WBC, age, and IgG were the top contributors to the model, while novel inflammatory indices were completely excluded. Conclusion Novel immune-inflammatory indices lack provide no independent discriminative value for moderate-to-severe SjD activity when adjusted for traditional parameters. The XGBoost machine learning model effectively handles non-linear variables and highlights the core diagnostic importance of patient-reported outcomes (ESSPRI) and traditional clinical features, providing a favorable evaluation tool. <Table Float="No" ID="Taba"> <tgroup cols="2"> <colspec align="left" colname="c1" colnum="1" /> <colspec align="left" colname="c2" colnum="2" /> <tbody> <row> <entry align="left" nameend="c2" namest="c1"> Key Points • Correlation analysis showed that traditional markers (ESR, CRP, IgG) and novel indices (NLR, CAR, CLR, IBI) were significantly associated with higher ESSDAI scores but showed weak-to-moderate linear correlations with nonlinear saturation trends. • In multivariable logistic regression, traditional markers remained independent risk factors, whereas novel indices lost significance. • The XGBoost model outperformed logistic regression and random forest, achieving a validation AUC of 0.701. • Feature importance identified patient-reported outcomes, HB, WBC, age, and IgG as top contributors, while novel indices were excluded. </entry> </row> </tbody> </tgroup> </Table>

Comparative Value of Traditional versus Novel Immune-Inflammatory Indices in Assessing Disease Activity of Sjögren’s Disease: A Multicenter Cohort Study

期刊论文
发表日期：

2026年7月14日

查看全文

Yaqi Zhao,
Xu Dong,
Pingting Yang,
Jiuliang Zhao,
Wei Bai,
Xinping Tian,
Wen Zhang,
Mengtao Li,
Xiaofeng Zeng,
Xiaomei Leng

摘要

Background

To evaluate the comparative value of traditional clinical features and novel immune-inflammatory indices in assessing disease activity in primary Sjögren's Disease (SjD), and to construct a robust classification model for moderate-to-severe disease activity using a machine learning approach.

Methods

This cross-sectional study included 18,738 SjD patients from the Chinese Rheumatism Data Center (CRDC) multicenter prospective cohort. Spearman rank correlation and local polynomial regression (LOESS) were used to explore the relationship between systemic inflammatory indices and the EULAR Sjögren's syndrome disease activity index (ESSDAI). To identify moderate-to-severe disease activity (ESSDAI ≥ 5), data were randomly split into training and validation sets (7:3). Classification models were developed using various machine learning techniques.

Results

Higher ESSDAI scores were significantly associated with elevated levels of both traditional markers (ESR, CRP, IgG) and novel composite indices (NLR, CAR, CLR, IBI) (all p < 0.001). However, these indices exhibited only weak-to-moderate linear correlations with ESSDAI and displayed non-linear saturation trends. In multivariable logistic regression, traditional markers remained independent risk factors, whereas all novel composite indices lost statistical significance, yielding a poor validation area under the curve (AUC) of 0.678. Conversely, the XGBoost model demonstrated modestly favorable overall performance Feature importance analysis revealed that patient-reported outcomes, hemoglobin, WBC, age, and IgG were the top contributors to the model, while novel inflammatory indices were completely excluded.

Conclusion

Novel immune-inflammatory indices lack provide no independent discriminative value for moderate-to-severe SjD activity when adjusted for traditional parameters. The XGBoost machine learning model effectively handles non-linear variables and highlights the core diagnostic importance of patient-reported outcomes (ESSPRI) and traditional clinical features, providing a favorable evaluation tool.

Key Points

• Correlation analysis showed that traditional markers (ESR, CRP, IgG) and novel indices (NLR, CAR, CLR, IBI) were significantly associated with higher ESSDAI scores but showed weak-to-moderate linear correlations with nonlinear saturation trends.

• In multivariable logistic regression, traditional markers remained independent risk factors, whereas novel indices lost significance.

• The XGBoost model outperformed logistic regression and random forest, achieving a validation AUC of 0.701.

• Feature importance identified patient-reported outcomes, HB, WBC, age, and IgG as top contributors, while novel indices were excluded.