Medication Extraction is the clinical NLP task of automatically identifying all medication entities and their associated attributes — drug name, dosage, route, frequency, duration, and indication — from clinical notes, discharge summaries, and patient records — forming the foundation of medication reconciliation systems, drug safety monitoring, and clinical decision support tools that depend on a complete and accurate medication list.

What Is Medication Extraction?

• Core Task: Named entity recognition targeting medication-related entities in clinical text.
• Entity Types: Drug Name (trade/generic), Dosage (amount + unit), Route (PO/IV/IM/SC/topical), Frequency (QD/BID/TID/QID/PRN), Duration, Reason/Indication.
• Key Benchmarks: i2b2/n2c2 2009 Medication Challenge, n2c2 2018 Track 2 (ADE and medication extraction), MTSamples dataset, SemEval-2020 Task 8.
• Normalization Target: Map extracted drug names to RxNorm, NDF-RT, or DrugBank identifiers for interoperability.

The i2b2 2009 Medication Challenge Format

The landmark benchmark. Input clinical note excerpt: "Patient was started on metformin 500mg PO BID with meals for newly diagnosed type 2 diabetes. Lisinopril 10mg daily was continued for hypertension. Patient reports taking ibuprofen 400mg PRN for joint pain."

Expected extractions:

Why Medication Extraction Is Hard

Non-standard Abbreviations: Clinical shorthand varies by institution, specialty, and individual clinician:

• "1 tab PO QHS" = 1 tablet by mouth at bedtime.
• "0.5mg/kg/day div q6h" = weight-based divided dosing — requires parsing mathematical expressions.
• "hold if SBP<90" = conditional dosing — medication held under hemodynamic condition.

Implicit Medications: "Continue home regimen" or "as previously prescribed" reference medications not explicitly named.

Negated Medications: "No anticoagulants" or "patient refuses insulin" — drug mention without active prescription.

Medication Changes: "Increased lisinopril to 20mg" vs. "decreased to 5mg" — dose change detection requires temporal comparison.

Polypharmacy Scale: Complex patients may have 15-30 medications across multiple specialty providers — extraction must be comprehensive with no omissions.

Performance Results

Clinical Applications

Medication Reconciliation:

• At transitions of care (ED to admission, admission to discharge), compile a complete medication list from all available notes.
• Prevents the ~40% medication discrepancy rate at hospital transitions that causes adverse events.

Drug Safety Alerts:

• Extract current medications as prerequisite for DDI screening.
• Alert prescribers when extracted medications interact with newly ordered drugs.

Polypharmacy Management:

• Population-level extraction identifies patients on high-risk medication combinations (≥5 medications, Beers Criteria drugs in elderly patients).

Research Data Extraction:

• Extract medication history for pharmacoepidemiology studies — which drugs were patients taking before their cancer diagnosis, cardiac event, or adverse outcome.

Medication Extraction is the medication safety foundation of clinical NLP — automatically compiling the complete, structured medication record from the free text of clinical documentation, enabling every downstream drug safety, interaction, and compliance application to operate on accurate, comprehensive medication data.