Objective To study the risk factors that can be used to distinguish children's refractory mycoplasma pneumonia (RMPP) from mycoplasma pneumonia (MPP),thus to determine the combined predictors for predicting RMPP,so as to provide basis for customizing accurate treatment and medication plan for RMPP children. Methods A total of 141 children with MPP admitted to Beijing Chaoyang Hospital Affiliated to Capital Medical University from January 1st,2016 to December 31st,2018 were included,and clinical data and laboratory examination data were collected. The patients were divided into MPP group and RMPP group according to the therapeutic efficacy of the first 3 days. The risk factors related to RMPP were preliminarily screened out by univariate analysis,and a multivariate Logistic regression model was established and independent risk factors were selected through stepwise regression. Finally,the receiver operating characteristic curve (ROC) of combined factors was drawn for prediction efficiency analysis. Results There were 85 cases of MPP and 56 cases of RMPP among 141 children.Age,pleural effusion,cough duration,C-creactive protein (CRP),erythrocyte sedimentation rate (ESR),lymphocyte count (LY)and percentage of lymphocytes (LY%) were significantly different between MPP group and RMPP group (P<0.05). Pleural effusion,fever duration,cough duration,CRP and LY% finally entered the prediction model of RMPP,with AUC of 0.834(95%CI:0.768—0.900),Youden index of 0.522 at the maximum. And the sensitivity and specificity were 80.4% and 71.8%,respectively. Conclusion This study constructs a combined model for predicting RMPP with pleural effusion,fever duration,cough duration,CRP,LY%,which has clinical guiding significance for the predictive diagnosis of RMPP and the use of glucocorticoids.