OBJECTIVE

2

To construct ensemble learning models for predicting high-use days of budesonide based on meteorological factors， thereby providing reference for hospital pharmacy management.

METHODS

2

Meteorological data for 2024 and outpatient budesonide usage data from the jurisdiction of Sanming Hospital of Integrated Traditional Chinese and Western Medicine were collected. High-use days were defined as the 75th percentile of outpatient budesonide usage， and a corresponding dataset was established. The prediction task was formulated as a classification problem， and three ensemble learning models were developed： Random Forest， Extreme Gradient Boosting （XGBoost）， and Histogram-based Gradient Boosting Classifier. Model performance was evaluated using accuracy， precision， recall， F1-score， and log-loss. Model interpretability was analyzed using Shapley Additive Explanations （SHAP）.

RESULTS

2

The Histogram-based Gradient Boosting Classifier achieved the best performance （accuracy＝0.75， F1-score＝0.48）， followed by XGBoost （accuracy＝0.74， F1-score＝0.43） and Random Forest （accuracy＝0.72， F1-score＝0.22）. SHAP results suggested that the prediction results of the last two models have the highest correction.

CONCLUSIONS

2

Ensemble learning models can effectively predict high-use days of budesonide， with the Histogram-based Gradient Boosting Classifier demonstrating the best predictive performance. Low temperature， high humidity， and low atmospheric pressure show significant positive impacts on the prediction of daily budesonide usage.