Flags for Facilitating Statistical Analysis Using CDISC Analysis Data Model
Chun Feng, Xiaopeng Li, Nancy Wang · 2013
CDISC Analysis Data Model (ADaM) is designed to be analysis-ready for statistical summary and analysis. Driven by analyses in Statistical Analysis Plan (SAP), correct analysis flags should be set up in ADaM data sets prior to analyses. The ADaM subject-level analysis data set (ADSL) contains subject-level population flags which indicate whether subjects are in efficacy, safety, pharmacokinetic, pharmacodynamic, food effect, and dose proportionality analyses. In Basic Data Structure (BDS) data sets, common record-level analysis flags include, for example, recheck flags, flags for exclusion, baseline flags, early termination identifiers, and treatment emergent flags. However, from our early phase clinical trial experience with ADaM, some analyses require additional customized analysis flags. For example, multiple baseline flags, average values as baselines flags, LOCF (last observation carried forward)/ WOCF (worst value carried forward) flags, extra treatment breakdown flags for adverse events summarization, and analysis flags for sub-group analysis. In the paper, we first categorize basic key analysis flags and summarize their functions in a checklist. In addition, we will share some examples with customized analysis flags used in our practice. We will also extend the topic to what to consider while setting up analysis-ready ADaM data. The paper will prepare ADaM data users (implementers and reviewers) to check analysis flags appropriate for intended analyses.