PyMAUDE: A Python Library for Local and Reproducible Analysis of the US Food and Drug Administration's Manufacturer and User Facility Device Experience (MAUDE) Database

The US Food and Drug Administration's (FDA) Manufacturer and User Facility Device Experience (MAUDE) database is the primary national repository of medical device adverse event reports.Although the FDA provides several access routes to MAUDE, each either limits query complexity or requires researchers to write study-specific code to process and query the data, complicating verification and reuse.Furthermore, the FDA does not release discrete versions of MAUDE and instead updates it in place and without version labels, making it difficult to link a given analysis to a specific state of the data.We developed PyMAUDE, a free and open-source Python package that builds a local, date-stamped copy of the MAUDE database in a few lines of code and detects revisions to previously downloaded FDA files by checksum.A built-in archiving function preserves the database as a set of Parquet files alongside a bundle of the raw source files, together suitable for upload to an online repository for later citation.PyMAUDE additionally provides advanced query capabilities; through Python function calls, users can match device fields, combine search term sets with AND/OR logic, or execute custom structured query language (SQL) queries.Retrieval from MAUDE was assessed by comparing the unique report keys of all 26.6 million reports in the database produced by PyMAUDE to the openFDA bulk export, finding that 26,557,462 (99.998%) of 26,558,073 unique report keys present in either source were present in both, with all 611 discrepant report keys confirmed to reflect differences in field content or presence of reports in the FDA's distribution channels rather than errors in PyMAUDE data processing.Across five exact-match queries returning 774 to 129,149 reports, PyMAUDE and the openFDA application programming interface (API) returned identical report-key sets in two cases and differed in each of the remaining three, with each difference attributable to PyMAUDE's case-insensitive matching discovering more records than the API or to data-channel differences.In our benchmarks on a consumer laptop, full database construction from pre-downloaded source files took a median of 9 minutes and 14 seconds, resulting in a database size of 57.7 GB on disk.Once built, exact-match queries on device fields each completed in less than 7.0 seconds, and ungrouped device-name searches combining terms with AND/OR logic each completed in less than 5.3 seconds.Archiving the complete database took a median of 3 minutes and 19 seconds, producing a 13.7 GB archive.PyMAUDE is available on GitHub and via PyPI under the MIT license, with source code and a prebuilt database snapshot as of October 21, 2026 archived on Zenodo.By pairing a programmatic query interface with a citable database snapshot, PyMAUDE enables device safety researchers to perform custom, complex, and reproducible analyses of MAUDE.