Frozen FDA MAUDE dataset and analysis code: comparative postmarket surveillance of ACL repair and reconstruction devices

This record contains the frozen FDA MAUDE dataset and the complete analysis code for a comparative postmarket surveillance study of anterior cruciate ligament (ACL) repair and reconstruction devices. It accompanies a manuscript comparing adverse event reporting for the Bridge-Enhanced ACL Restoration (BEAR) Implant against soft-tissue fixation devices reported in ACL procedures. Two cohorts were retrieved from the openFDA device event endpoint (https://api.fda.gov/device/event.json) between 01:29:51 and 01:30:26 UTC on 3 August 2026. The index cohort is every medical device report filed under product code QNI, the single-device product class created for the BEAR Implant by De Novo DEN200035, giving 35 reports and therefore a census rather than a sample. The comparator cohort is 2,196 reports filed under product codes MBI and MAI whose narrative text contains the phrase "anterior cruciate". Both cohorts begin 1 December 2020, the month of BEAR authorization. A text filter was necessary because PPQ, the only ACL-specific product code for conventional reconstruction, has no medical device reports in the entire history of the database; that absence is itself a reported finding. The archive contains the frozen cohort CSVs, the raw openFDA JSON responses, a retrieval manifest recording the exact query strings, the SHA-256 of the retrieval script and the UTC timestamps, the study protocol and its amendments, the failure-mode codebook, the dual-coding extraction instrument, and every analysis script. Running the scripts against the frozen data reproduces each table and figure in the manuscript from a clean state. MAUDE records are edited retroactively and reports are added for events that occurred years earlier, so re-querying openFDA will not reproduce these results. The deposited snapshot is the reference dataset, and the manifest is what makes the difference auditable. The comparator text filter was validated on a random sample of 100 reports (seed 20260803) and correctly identified an ACL-related procedure in 98, giving a positive predictive value of 98.0 percent (Wilson 95% CI 93.0 to 99.4). A secondary variable recording device role was dual-coded; agreement was 97.0 percent with Cohen's kappa 0.880 before reconciliation, and the pre-reconciliation labels are preserved alongside the adjudicated ones. After reconciliation, 15 of the 100 sampled comparator reports describe hardware deployed during an ACL procedure without serving an ACL fixation role, so the comparator is best described as soft-tissue fixation devices reported in ACL procedures rather than as conventional ACL reconstruction hardware. One disclosure. The initial pass of the text-filter validation was performed by a language model rather than a human reviewer. This is stated in the protocol, recorded in the workbook itself, and reported in the manuscript. The human review is the figure carried in the paper; the machine pass is retained only as a comparison