AUTHOR=Zhai Xvwen , Feng Min , Guo Hui , Liang Zhaojun , Wang Yanlin , Qin Yan , Wu Yanyao , Zhao Xiangcong , Gao Chong , Luo Jing TITLE=Development of Prediction Models for New Integrated Models and a Bioscore System to Identify Bacterial Infections in Systemic Lupus Erythematosus JOURNAL=Frontiers in Cellular and Infection Microbiology VOLUME=11 YEAR=2021 URL=https://www.frontiersin.org/journals/cellular-and-infection-microbiology/articles/10.3389/fcimb.2021.620372 DOI=10.3389/fcimb.2021.620372 ISSN=2235-2988 ABSTRACT=Objectives

Distinguishing flares from bacterial infections in systemic lupus erythematosus (SLE) patients remains a challenge. This study aimed to build a model, using multiple blood cells and plasma indicators, to improve the identification of bacterial infections in SLE.

Design

Building PLS-DA/OPLS-DA models and a bioscore system to distinguish bacterial infections from lupus flares in SLE.

Setting

Department of Rheumatology of the Second Hospital of Shanxi Medical University.

Participants

SLE patients with flares (n = 142) or bacterial infections (n = 106) were recruited in this retrospective study.

Outcome

The peripheral blood of these patients was collected by the experimenter to measure the levels of routine examination indicators, immune cells, and cytokines. PLS-DA/OPLS-DA models and a bioscore system were established.

Results

Both PLS-DA (R2Y = 0.953, Q2 = 0.931) and OPLS-DA (R2Y = 0.953, Q2 = 0.942) models could clearly identify bacterial infections in SLE. The white blood cell (WBC), neutrophile granulocyte (NEUT), erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), procalcitonin (PCT), interleukin-6 (IL-6), IL-10, interferon-γ (IFN-γ), and tumor necrosis factor α (TNF-α) levels were significantly higher in bacteria-infected patients, while regulatory T (Treg) cells obviously decreased. A multivariate analysis using the above 10 dichotomized indicators, based on the cut-off value of their respective ROC curve, was established to screen out the independent predictors and calculate their weights to build a bioscore system, which exhibited a strong diagnosis ability (AUC = 0.842, 95% CI 0.794–0.891). The bioscore system showed that 0 and 100% of SLE patients with scores of 0 and 8–10, respectively, were infected with bacteria. The higher the score, the greater the likelihood of bacterial infections in SLE.

Conclusions

The PLS-DA/OPLS-DA models, including the above biomarkers, showed a strong predictive ability for bacterial infections in SLE. Combining WBC, NEUT, CRP, PCT, IL-6, and IFN-γ in a bioscore system may result in faster prediction of bacterial infections in SLE and may guide toward a more appropriate, timely treatment for SLE.