Representation and Extraction of Drug Information: A Study Focusing on the Needs of Older Adults

Poster Number: P186
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Information Extraction, Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

As the population ages, older adults (roughly those aged 65+) face the challenge of managing complex medication
regimens, requiring significant information about the drugs they are taking. However, supporting medication information technologies like alerts, apps, and chatbots face hurdles due to the unstructured nature of the authoritative source of drug information–FDA Structured Product Labels, commonly known as “drug labels”–which are almost entirely free-text documents. This paper investigates the effectiveness of using natural language processing (NLP) to extract critical drug information from these documents using a small set of 20 drug labels annotated with drug indications, contraindications, adverse reactions, dosages, and usage instructions. The study employs (and makes publicly available) a BERT model pre-trained on 122 thousand drug labels. We also explored transfer learning using an existing adverse reaction dataset. Despite the limited size of our pilot data, resulting in a promising performance, including an F1 of 94 for adverse reactions, 80 for dosages, 90 for instructions, and 76 for contraindications, though only 52 for indications. With further development, this line of research can lead to enhanced medication management for older adults.

Speaker(s):
Abayomi Adegunlehin, MS
University Of Texas Health Science Center at Houston

Author(s):
Kirk Roberts, PhD - University of Texas Health Science Center at Houston; Francis Ifiora, MS - University of Texas Health Science Center at Houston; Tasneem Kaochar, MS - University of Texas Health Science Center at Houston;