A combined risk model for the multi-encompassing identification of heterogeneities of prognoses, biological pathway variations and immune states for sepsis patients

BMC Anesthesiology

Guan-hua Li¹  Shi Zhang¹  Rui Zang¹  Long-bin Pang¹  Jing Wang¹  Zong-xiu Yin¹  Chun-yan Xing¹  Jing Pan¹ 

[1] Department of Pulmonary and Critical Care Medicine, Jinan Central Hospital Affiliated to Shandong First Medical University and Shandong Academy of Medical Sciences, Cheeloo College of Medicine, Shandong University, No. 105 Jiefang Road, 250013, Jinan, Shandong Province, China;Department of Pulmonary and Critical Care Medicine, Jinan Central Hospital Affiliated to Shandong First Medical University, Jinan, China;

关键词: Sepsis;    Heterogeneity;    Prediction model;    Signal transduction pathway;    Immune dysfunction;   

DOI  :  10.1186/s12871-021-01552-x

来源: Springer

【 摘 要 】

BackgroundSepsis is a highly heterogeneous syndrome with stratified severity levels and immune states. Even in patients with similar clinical appearances, the underlying signal transduction pathways are significantly different. To identify the heterogeneities of sepsis from multiple angles, we aimed to establish a combined risk model including the molecular risk score for rapid mortality prediction, pathway risk score for the identification of biological pathway variations, and immunity risk score for guidance with immune-modulation therapy.MethodsWe systematically searched and screened the mRNA expression profiles of patients with sepsis in the Gene Expression Omnibus public database. The screened datasets were divided into a training cohort and a validation cohort. In the training cohort, authentic prognostic predictor characteristics (differentially expressed mRNAs, pathway activity variations and immune cells) were screened for model construction through bioinformatics analysis and univariate Cox regression, and a P value less than 0.05 of univariate Cox regression on 28-day mortality was set as the cut-off value. The combined risk model was finally established by the decision tree algorithm. In the validation cohort, the model performance was assessed and validated by C statistics and the area under the receiver operating characteristic curve (AUC). Additionally, the current models were further compared in clinical value with traditional indicators, including procalcitonin (PCT) and interleukin-8 (IL-8).ResultsDatasets from two sepsis cohort studies with a total of 585 consecutive sepsis patients admitted to two intensive care units were downloaded as the training cohort (n = 479) and external validation cohort (n = 106). In the training cohort, 15 molecules, 20 pathways and 4 immune cells were eventually enrolled in model construction. These prognostic factors mainly reflected hypoxia, cellular injury, metabolic disorders and immune dysregulation in sepsis patients. In the validation cohort, the AUCs of the molecular model, pathway model, immune model, and combined model were 0.81, 0.82, 0.62 and 0.873, respectively. The AUCs of the traditional biomarkers (PCT and IL-8) were 0.565 and 0.585, respectively. The survival analysis indicated that patients in the high-risk group identified by models in the current study had a poor prognosis (P < 0.05). The above results indicated that the models in this study are all superior to the traditional biomarkers for the predicting the prognosis of sepsis patients. Furthermore, the current study provides some therapeutic recommendations for patients with high risk scores identified by the three submodels.ConclusionsIn summary, the present study provides opportunities for bedside tests that could quantitatively and rapidly measure heterogeneous prognosis, underlying biological pathway variations and immune dysfunction in sepsis patients. Further therapeutic recommendations for patients with high risk scores could improve the therapeutic system for sepsis.

【 授权许可】

CC BY