Unveiling psychobiological correlates in primary Sjögren’s syndrome: a machine learning approach to determinants of disease burden

TL;DR

This study explores how psychological and biological factors together influence disease burden in primary Sjögren’s syndrome using machine learning.

Contribution

The study quantitatively compares biopsychosocial factors in pSS using machine learning to identify novel predictors of disease burden.

Findings

Trait anxiety was a significant negative predictor of autoantibodies and ESSPRI.

Biological markers like IgG and RF were important predictors of disease burden.

Psychological traits like 'Fatigability' and 'Pure-hearted conscience' showed significant predictive power for ESSPRI.

Abstract

Besides primary Sjögren’s syndrome (pSS) is generally assessed through biological markers, growing evidence suggests that psychological and social factors—such as anxiety, depression, personality traits, and social support—may also play a role in disease burden. Relative contribution of these biopsychosocial dimensions to disease activity in pSS, however, has not been quantitatively compared. This study aimed to evaluate the predictive weight of different factors in determining both objective and subjective disease burden using machine learning (ML) models.

117 pSS patients, whose biological (blood cell counts, complement activity, IgG, RF, SSA, SSB), psychological (personality traits, depression, anxiety, basic self-esteem assessed via self-reported questionnaires), and social (socioeconomic status and social support) measures were collected in a composite database. Outcome variables were SSA/SSB autoantibodies and EULAR Sjögren Syndrome Patient Reported Index (ESSPRI), as indicators of biological and perceived disease burden, respectively. Three machine learning algorithms were trained to predict outcome variables, first by each measure category, then on the entire set of predictor variables. Permutation feature importance was used to assess the importance of the predictors. The five most important predictors were selected for all target outcomes.

Concerning autoantibodies, the model performed best with biological input only, in the case of ESSPRI, the complete dataset gave the best performance. Trait anxiety was selected as important negative predictor of both autoantibodies. Besides, biological measures (IgG, RF, platelet count) and age were among the five most important features. State anxiety and temperament trait ‘Fatigability’ were important positive predictors of ESSPRI, while character trait ‘Pure-hearted conscience’, IgG and RF were important negative predictors.

Unexpected psychobiological correlations, like trait anxiety and IgG/RF as negative predictors of autoantibodies and ESSPRI, respectively, suggest different (immunobiological and psychosomatic) disease mechanisms and symptom burden. Importance of psychological factors in estimating disease burden may pave the way toward novel, more sensitive diagnostic tools and therapeutic methods and better understanding of pathomechanisms of pSS.

Figures