Replicating medication trend studies using ad hoc information extraction in a clinical data warehouse.

Area of Science:

• Health Informatics
• Pharmacovigilance
• Clinical Data Management

Background:

• Electronic Health Records (EHR) often store crucial patient data, including medication information, as unstructured free text.
• Conventional information extraction (IE) methods require significant development effort for data retrieval.
• Ad hoc IE offers a more efficient alternative for querying and extracting on-the-fly information from clinical Data Warehouses (CDW).

Purpose of the Study:

• To develop and validate a generalizable ad hoc IE approach for extracting pharmacotherapy details from hospital discharge letters.
• To enhance CDW systems with features supporting accurate medication and dosage extraction, including error tolerance and proximity search.
• To facilitate the replication of medication trend studies using real-world clinical data.

Main Methods:

• Implemented ad hoc IE techniques within a CDW, incorporating features for error-tolerant queries and proximity searches.
• Filtered out negated, historical, and non-patient-specific data during the CDW integration process.
• Utilized Anatomical Therapeutic Chemical (ATC) classification system drug lists to query medication data.

Main Results:

• Achieved high F1 scores: 0.983 for medication extraction and 0.974 for daily dosage extraction from discharge letters.
• Successfully replicated 93% of main findings and 75% of all findings across three published medication trend studies (hypertension, atrial fibrillation, chronic kidney disease).
• One study was completely replicated, demonstrating the approach's effectiveness.

Conclusions:

• Ad hoc IE provides a novel, efficient method for extracting medication and dosage information from clinical texts like discharge letters.
• This approach significantly enhances CDW search capabilities beyond conventional methods.
• It enables fast and high-quality replication of medication trend studies, supporting pharmacotherapy research.