Medication Extraction and Reconciliation Knowledge Instrument — extract structured medication information (drug, dose, route, frequency, prn, dates) from free-text clinical notes.
MERKI parses narrative clinical text and pulls out medications. Each result is either a drug (the name was found in the bundled RxNorm-derived lexicon) or a possible drug (the name wasn’t in the lexicon — e.g. a misspelling — but it was flanked by enough evidence, like a dose and instructions, a “prn”, or a “treated with …” phrase, to be worth surfacing). The parser also tracks context clues (“Discharge medications:”, “Allergies:”, “Home meds:”, “on hold”, …): a context applies to every medication after it until the next context clue.
For each medication the table shows the drug name and any dose, route, frequency, prn, and date it found, plus
the inferred when (before admission / during stay / after discharge) and the surrounding text with
the match highlighted. The raw token objects are under “Show JSON”.
MERKI was built at Columbia University’s Department of Biomedical Informatics. The original paper — Extracting Structured Medication Event Information from Discharge Summaries (Gold, Elhadad, Zhu, Cimino, Hripcsak) — received an AMIA Distinguished Paper award. This page is a dependency-light TypeScript port of the original Perl parser, so the demo can run as a static page; the original Perl-CGI demo is long gone. Source and the parser library are in the repository (GPL-3.0).