Identifying Medication Discontinuations Using Text Searching and Clinical Data

Area of Science:

• Pharmacology
• Health Informatics
• Geriatrics

Background:

• Identifying clinically intended medication discontinuations is crucial for deprescribing research.
• Existing methods may misclassify discontinuations, hindering evidence generation.

Purpose of the Study:

• To develop and validate an algorithmic approach for identifying clinically intended medication discontinuations at scale.
• To combine text strings from clinical documentation with medication order data for improved accuracy.

Main Methods:

• A cohort of 1588 individuals over 65 with medication gaps was manually reviewed to establish a gold standard.
• Text strings were developed to query clinical documentation for discontinuation intent, tailored to specific drug classes.
• The algorithm integrated text data with medication order discontinuation data.

Main Results:

• The full algorithm demonstrated moderate to high sensitivity and specificity across five drug groups (oral hypoglycemics, statins, antihypertensives, bladder antimuscarinics, antithrombotics).
• Performance metrics included 80% sensitivity/85% specificity for oral hypoglycemics and 75% sensitivity/95% specificity for statins.
• The combined algorithm outperformed text or order data alone in identifying intended discontinuations.

Conclusions:

• Text-based approaches, enhanced by medication order data, can identify clinically intended medication discontinuations with acceptable accuracy.
• This algorithmic method offers a potential improvement over claims data for reducing misclassification in deprescribing studies.