Abstract:
|
Safety monitoring is an important area in clinical development. To better characterize the safety profile with clinical trial data, data science technique emerges as an adverse event detection and risk assessment tool. However, little attention has been draw to how such methods could be used in safety evaluation. In this work, we specifically focus on how discriminant analysis can be used to identify the higher risk groups and predict the adverse event of interest. Starting from the widely used linear discriminant analysis (LDA), Bayesian discriminant analysis (BDA) is also utilized as prior knowledge could be helpful for safety evaluation. Furthermore, we extended the LDA method to establish a relationship between the patient information and occurring of rare events. Three approaches for rare event classification are compared in this work. Simulations are conducted to compare different methods. The performance of the compared methods is also examined through a real case study through an AIDs clinical trial data.
|