Abstract
BACKGROUND: Safety data is required to be collected in all clinical trials and can be separated into two types of data, adverse events and serious adverse events. Often, these types of safety data are collected as two discrete data sets, where adverse events that also meet the criteria for seriousness should be reported in both datasets. Safety analyses are often conducted using only the adverse event dataset, which should feature all safety events reported. We investigated whether the reporting of safety in both datasets was systematically followed and explored the impact of this on safety analyses in ICON8, an ovarian cancer clinical trial. METHODS: Text searches of serious adverse event data identified events that could potentially match the data reported in the adverse event dataset (looking at pre-specified AE terms only). These serious adverse events were then mapped to adverse event data according to predefined criteria: (a) event term matches, (b) date of onset and date of assessment within 30 days of each other, (c) date of assessment lies between date of onset and date of resolution and (d) events confirmed to occur in the same chemotherapy cycle. A combined dataset of all unique safety events (whether originally reported in the adverse event or serious adverse event dataset) was created and safety analyses re-performed. RESULTS: 51,019 adverse events were reported in ICON8, of which 42,410 were included in the mapping exercise. One thousand five hundred six serious adverse event elements were reported, of which 668 were included in the mapping exercise. Sixty-one percent of serious adverse event elements was matched to an already-reported adverse event. Supplementing these additional safety events and re-performing safety analyses increased the proportion of patients with at least one grade 3 or worse safety events in all arms from 42 to 47% in the control arm and 61 to 65% and 52 to 59% in the research arms. The difference in proportions of grade 3 or worse event in the research arms compared to the control arm changed by 18% (95% confidence interval [CI] 12 to 24%) and 12% (95% CI 6 to 18%), respectively. CONCLUSIONS: There was low agreement in mapping serious adverse events to already reported adverse events, with nearly 40% of serious adverse events included in the mapping exercise not mapped to an already reported adverse event. Any analyses of safety data that use only adverse event datasets or do not clearly account for serious adverse event data will likely be missing important safety information. Reporting standards should make clear which datasets were used for analyses.