Analysis data and code for "Approvals are not independent observations: within-molecule comparison of primary endpoints across 450 oncology approval events, 2015-2026"

Data and code for a methodological study of how two analytic choices affect a within-molecule comparison of primary endpoints across US oncology approval events, 2015 to 2026. The two choices are the unit of analysis and the rule used to collapse repeated approvals of the same molecule. The deposit contains the event-level table on which every estimate rests (p1a_analysis_set.csv, 450 approval events, 39 fields), the adjudication record used for the endpoint-ascertainment check (p1a_adjudication.csv, 107 reviewed events), five analysis scripts in code/, the four JSON result files from which the manuscript tables are built, and Figures 1 to 3 in figures/, as PDF and as 600 dpi PNG. README.md documents every field, the label-version selection rule and the command that regenerates each result. LICENSE.txt gives the terms: data and results under CC BY 4.0, code under MIT.The record licence is CC BY 4.0 and covers every file, so the code can be reused under either. All inputs are public. Approval actions and submission types come from the Drugs@FDA submissions file distributed through openFDA. Active ingredients were classified to ATC groups L01 and L02 through RxNorm and RxClass. The primary endpoint designated at each action was read from the Clinical Studies section of the product label in force at the action date, retrieved from DailyMed. Python 3.11 with pandas, numpy, scipy, statsmodels and matplotlib reproduces every reported estimate, and the included JSON files were produced by these scripts against these CSV files. A full rerun takes a few minutes. The Clinical Studies text itself is not redistributed, because it is third-party content from FDA-approved labelling; one anchor sentence per adjudicated event, capped at 400 characters, is retained, and the submission and label identifiers in the analysis set are sufficient to retrieve the source document for any event. The deposit accompanies the manuscript and the analysis plan registered on OSF.