This study proposes a decision support system for drug therapy in the ICU based on dynamic adaptive generative models. The system is designed with a layered architecture, including a data layer, model layer, and application layer, utilizing the Dynamic Adaptive Generative Polynomial Regression (DAGPR) algorithm to analyze and model medication data for ICU patients. The data layer implements data preprocessing and feature engineering, while the model layer employs dynamic adaptive strategies for model training and optimization. The application layer constructs an interactive interface using the Streamlit framework, allowing healthcare personnel to upload and analyze patient data in real time. Experimental results demonstrate that the system outperforms traditional empirical medication methods in terms of drug accuracy, response time, and predictive accuracy, effectively shortening treatment duration, improving patient outcomes, and offering significant economic benefits. The research findings provide reliable technical support for precision medication decisions in the ICU, showcasing good clinical application value.