Nomogram construction based on characteristic genes and clinical variables to predict the risk of multiple organ dysfunction syndrome caused by influenza in children.

Abstract

Background: Screening for risk factors for the occurrence of multiple organ dysfunction syndrome (MODS) caused by pediatric influenza is an essential approach to improving treatment interventions and stratifying prognosis. This study aimed to select characteristic genes in MODS samples, demonstrate the correlation between characteristic genes and clinical variables, show the changes in expression levels of characteristic genes in the progression of MODS, and establish a predictive prolonged MODS (PM) line chart model.

Methods: We downloaded the pediatric influenza blood messenger ribonucleic acid (mRNA) dataset (GSE236877) from the Gene Expression Omnibus (GEO) database. Multiple logistic regression analyses were employed to screen for risk factors and independent risk factors, and to establish nomogram model. The receiver operating characteristic (ROC) curve was used to evaluate the predictive efficacy of variables on disease occurrence, where a larger area under the curve (AUC) indicates better predictive performance. Calibration curves and the Hosmer-Lemeshow goodness-of-fit test were utilized to describe whether the curves exhibited deviation. Decision curve analysis (DCA) was employed to assess the predictive efficacy of the model.

Results: SLC12A7 was an independent risk factor that increased the risk of PM (OR =0.356, P<0.001). GNA15 (OR =4.598, P<0.001) and EMP1 (OR =2.158, P=0.002) were protective factors that reduced the risk of PM occurrence. These three genes were combined with clinical variables, including age, influenza virus type, and bacterial co-infection, to construct a nomogram model for predicting the risk of MODS in children with influenza. The AUC of the nomogram score was 0.946, which was larger than the AUC of individual genes and clinical variables. Nomogram model can increase the net benefit of patients compared with clinical variables.

Conclusions: TGFBI, SLC12A7, LY86, HAL, CASP5, RETN, ESPL1, TULP2, DEFB114, EMP1, GNA15, GPAA1 were characteristic genes that distinguished between never MODS (NM) and PM samples. SLC12A7, GNA15, and EMP1 can serve as independent predictive factors for MODS. A nomogram model based on SLC12A7, GNA15, EMP1, and clinical variables (age, influenza virus type, and bacterial co-infection status) demonstrated better predictive performance for the risk of MODS in children with influenza compared to clinical variables and single genes.