A Learning and Decision-Making Approach
PhRMA SPERT proposed three tiers (classifying endpoints by analysis) for analyzing AEs.
- Tier 1: Pre-specified events (p-values)
- Tier 2: More frequent events that are not pre-specified (confidence intervals)
- Tier 3: Less frequent events that are not pre-specified (descriptive statistics)
An alternative approach is to think of two parts (classifying endpoints by clinical interest) for assessing AEs and other safety data.
- Part 1 (for learning): All events are summarized in the overall safety assessment with descriptive statistics and graphical displays (CIs may be provided but no inferential statistics are included)
- Part 2 (for decision-making): Safety topics of interest are explored using more in-depth analyses and/or specific groupings of events that help to further characterize their occurrence (p-values are only provided for safety endpoints with explicit hypotheses)
On the surface this may appear to be a subtle difference; however, it fundamentally transforms a testing and confirming approach (adhering to an efficacy mindset with implicit hypothesis tests) into a learning and decision-making approach (a scientific evaluation of the safety data).