IDENTIFYING ALCOHOL - RELATED INFORMATION FROM UNSTRUCTURED BILINGUAL CLINICAL NOTES WITH MULTILINGUAL TRANSFORMS

Abstract

This study investigates the efficacy of multilingual transforms in identifying
alcohol-related information from unstructured bilingual clinical notes. With the
increasing importance of extracting meaningful insights from electronic health
records (EHRs), particularly in diverse linguistic contexts, we propose a novel
approach that leverages multilingual embeddings to capture the nuances of language
and improve the performance of alcohol-related information extraction. Our
methodology involves preprocessing bilingual clinical notes, applying multilingual
transforms to represent the text in a shared embedding space, and training a classifier
for alcohol-related information detection. We evaluate our approach on a dataset of
bilingual clinical notes, achieving promising results in identifying mentions of alcohol
use, abuse, and related health conditions across languages. The findings suggest that
multilingual transforms offer a robust solution for cross-lingual information extraction
tasks in healthcare, with potential applications in improving clinical decision support
systems and population health management strategies.