Brigham Young University

Author Of 1 Presentation

Disease Modifying Therapies – Risk Management Poster Presentation

P0296 - Automated extraction of multiple sclerosis treatment timelines (ID 1859)

Speakers
Presentation Number
P0296
Presentation Topic
Disease Modifying Therapies – Risk Management

Abstract

Background

Background: The treatment information stored in free-text electronic health records has remarkable potential for use in a variety of research applications, including pharmacogenomic studies investigating genetic variants associated with adverse drug reactions to multiple sclerosis (MS) medications. However, utilization of this data typically requires extensive manual curation.

Objectives

Objectives: To determine the feasibiity of extracting glatiramer acetate (GA) treatment timelines in an automated manner from electroni health records, to increase datasets available for MS treatment research.

Methods

Methods: We identified 7,000 patients with MS from a de-identified electronic health record database from Vanderbilt University Medical Center. After extraction of the medication recoreds, we developed a rule-based algorithm using Python and Sequel for GA that 1) classifies free-text notes mentioning GA based on patient treatment status and 2) aggregates note classifications into complete GA treatment periods.

Results

Results: For notes indicating GA initiation, continuation, and discontinuance, the algorithm achieved sensitivities of 0.79, 0.83, and 0.79, respectively. For these same groups, precision was 0.92, 0.97, and 0.65, respectively. Subsequently, the algorithm correctly identified 72% of treatment period beginnings (precision = 0.75) and 70% of treatment period ends (precision = 0.73). The algorithm correctly determined 86% of days during which patients were on GA (precision = 0.79). A conservative method of treatment period generation was developed that extracted fewer periods (per day sensitivity = 0.40) but maintained extremely high fidelity (per day precision = 0.98).

Conclusions

Conclusions: Automated extraction of MS treatment timelines is feasilbe with high accuracy. Among other applications, the treatment data extracted by the algorithm can facilitate pharmacogenetic research either as direct input in analyses or by significantly reducing the time demands of manual curation. Application of algorithms such as the one we have developed will increase the ability to study MS treatments in real world populations.

Collapse