Analysis of risk factors for primary Sjögren's syndrome complicated with hematological involvement

Objective To analyze clinical characteristics of patients with primary Sjögren's syndrome (pSS) complicated with hematological damage and to established the prediction models.

Methods The clinical characteristics and laboratory tests imaging examinations of 183 pSS patients were collected. A comparative analysis of pSS patients with hematological damage (n=109) or without hematological damage(n=74)was performed. Logistic regression was used to analyze the related factors of patients with pSS complicated with blood system damage. Prediction models of primary Sjögren's syndrome with hematological involvement based on artificial neural network (ANN) and multi-Logistic regression (MLR) were established, the receiver operating characteristic curve was drawn; and the area under the curve (AUC) and the basic parameters of the two models were compared to evaluate their predictive effectiveness of the models.

Results Logistic regression analysis showed that liver involvement (OR=0.191, P < 0.05, 95%CI, 0.070 to 0.524), skin involvement (OR=0.292, P < 0.05, 95%CI, 0.121 to 0.704), hypoalbuminemia (OR=0.840, P < 0.05, 95%CI, 0.743 to 0.948), hypokalemia (OR=0.351, P < 0.05, 95%CI, 0.145 to 0.846), high immune globulin (IgM)(OR=1.732, P < 0.05, 95%CI, 1.085 to 2.765), high erythrocyte sedimentation rate (ESR) (OR=1.028, P < 0.05, 95%CI, 1.005 to 1.051) and positive anti-SSA antibodies (OR=0.21, P < 0.05, 95%CI, 0.052 to 0.828)were risk factors associated with hematologic involvement in patients with pSS. There was no significant difference between MLR and ANN models in predicting platelet and leukocyte decline (P>0.05). ANN model was better than MLR model in predicting hemoglobin decline, and the difference was statistically significant (P < 0.05).

Conclusion Patients with pSS are more likely to suffer hematologic damage. Dysfunction of liver, skin involvement, hypoalbuminemia, hypokalemia, high IgM, high ESR and positive anti-SSA antibodies in clinical works are risk factors for blood system involvement in patients with pSS. ANN and MLR models are two validated predictive modeling tools developed based on the original data from this study, and could be considered for clinical application.