Construction of random forest prediction model and analysis of decision curve for refractory Mycoplasma pneumoniae pneumonia based on serum inflammatory factors

Objective To construct a random forest prediction model of refractory Mycoplasma pneumoniae pneumonia (RMPP) based on serum inflammatory factors, and to evaluate the prediction model through decision curve analysis. Methods A total of 990 children with Mycoplasma pneumoniae pneumonia (MPP) who were admitted to in the Maternal and Child Health Hospital of Shijiazhuang from January 2021 to February 2023 were included, and their clinical data and levels of serum inflammatory factors were collected. The children were randomly divided into two parts: a training set (n=693) and a verification set (n=297) at a ratio of 7∶3 using the sample software package of R4.1.3 software. With R4.1.3, the training data were divided into two parts:RMPP and general Mycoplasma pneumoniae pneumonia (GMPP) (GMPP=0, RMPP=1). Based on the random forest algorithm, the independent variables in the training set data were sorted by their feature importance, and the optimal variable combination was selected by VIMP combined with the minimum depth method to construct a random forest prediction model of RMPP. The prediction model was evaluated by verification set and decision curve analysis. Results The optimal variable combination of random forest prediction models for RMPP screened by random forest algorithm were interleukin (IL)-6,D-dimer (DD), lactate dehydrogenase (LDH), IL-10. Decision curve analysis showed that MPP children may benefit the most from clinical intervention after the threshold probability reached 6%. Conclusions The optimal variable combination of random forest prediction models for RMPP screened by random forest algorithm are IL-6, DD, LDH,IL-10. The above random forest prediction model for RMPP has good prediction efficiency.