Enhancing Adverse Event Monitoring and Management in Phase IV Chronic Disease Drug Trials: Applications of Machine Learning

At present, Phase IV studies usually rely on RWD and RWE to determine the long-term safety, effectiveness, and, cost benefits of drugs. This is applicable to public health and medical practices and impacts drug and health decision-making at the clinical level. In this paper, we firstly present a novel Transformer model-based method which could further promote the long-term application benefits of the chronic disease drugs including the metreleptin in lipodystrophy. Through the self-attention mechanism, the model could interestingly capture the time series and correlation in clinical data, and process the multimodal real world data (e.g., patient history, long-term follow up data, etc.), and is capable of real-time monitoring of drug safety and efficacy. In fact, due to the addition of data augmentation and self-supervised learning strategy, the model can still achieve high prediction performance in low sample settings, and quickly and accurately identify potential adverse events including the acute pancreatitis, liver adverse events, etc. and their role in glycosylated hemoglobin (HbA1c), triglyceride and other indicators. Experimental results demonstrate that the proposed Transformer model significantly outperforms traditional methods in adverse event prediction and risk assessment, making it a valuable tool for accurate long-term drug monitoring.