<b>Study of the best prediction model for the effectiveness of subcutaneous immunotherapy in childhood asthma patients</b>

<b>Objective:</b> To explore the best prediction model for the effectiveness of subcutaneous specific immunotherapy (SCIT) and the best time to achieve good disease control. <b>Methods: </b>A total of 141 children aged 5-14 years were diagnosed with allergic asthma with or without allergic rhinitis from January 2008 to June 2023. Dust mite allergy was determined by allergen examination and a controlled study. The patients were divided into withdrawal and nonwithdrawal groups according to treatment efficacy. Baseline data, conventional ventilated lung function, the daily drug score (DMS) and the visual analog scale (VAS) score were obtained. Variables with significant differences were included in the logistic regression model, random decision forest model and XGBoost model, and the model prediction ability and the area under the curve (AUC) were tested to determine the best model and the best time to achieve good disease control. <b>Results:</b> 1: There were 109 patients (77%) in the withdrawal group and 32 patients (23%) in the nondiscontinuation group; DMSs varied significantly at the sixth month, ninth month, 1st year, 2nd year, 3rd year, end of treatment (at the end of SCIT at and over 3 years), and the first year after the end of treatment. These results are conducive to drug withdrawal. Compared with those at treatment initiation, a decrease in the VAS score at the ninth month and the FEF 25% at the first year had an impact on withdrawal. 2. The AUCs of the logistic regression model, the random forest model, and the XGBoost model were 0.865, 0.9, and 0.84, respectively. Ten cross-validations were used to compare the AUCs of the three models for the prediction of the data set. 3. The average accuracy reduction of the random decision forest model at the 9th month for DMS (mean decrease in accuracy, MDA) was 11.77, and the model importance for VAS score in 9th month in the XG-Boost model was 0.121. <b>Conclusions</b>: 1. The random decision forest model is the best prediction model the effectiveness of SCIT in children with asthma. 2. Good disease control should be achieved in the ninth month of SCIT. 3. Allergy immunotherapy (AIT) for 3 years or more is the most beneficial for drug withdrawal.