Statistical analysis on adverse events of clinical trials

Abstract

Clinical trials evaluate the safety and efficacy of a drug. One of the methods used to evaluate patient safety within a clinical trial is to monitor and report all adverse events. Generally, the crude percentage (rate) of adverse events are tabulated and listed in the statistical reports of clinical trials. ☐ In this thesis, alternative methods for summarization and characterize of this type of data are discussed. Four attributes of adverse event are: occurrence/non-occurrence, severity, onset and duration of adverse events. The exact tests will characterize the occurrence/non-occurrence of AE. Nonparametric tests (Kaplan-Meier curves and log-rank test) for group comparisons of time to first adverse events and duration of AE. The model-based methods will include logistic regression, decision tree and random forest for dichotomous response variables, cox model regression and accelerative failure time model for time to onset of the adverse events. ☐ The analysis was using the adverse event dataset and subject level analysis dataset from CDISC - FDA Pilot Data Sets.