Clinical Report: Analyzing and Categorizing Medication Discontinuations in Estonia
Overview
This study utilized large language models (LLMs) to analyze medication discontinuations among Estonian patients, focusing on antidiabetics and statins. The research aimed to identify reasons for discontinuation and whether the decision was made by patients or physicians, providing insights into chronic disease management.
Background
Medication adherence is crucial for managing chronic diseases, impacting patient outcomes and healthcare costs. Discontinuation of medications is common and can arise from various factors, yet these reasons are often not systematically recorded in electronic health records (EHRs). This study addresses the gap in understanding medication discontinuation by employing advanced natural language processing techniques to analyze clinical narratives.
Data Highlights
The dataset comprised a 10% random sample of the Estonian population (n=150,824) from 2012 to 2019, focusing on antidiabetics and statins.
Key Findings
LLMs were effective in identifying and categorizing reasons for medication discontinuation from clinical narratives.
Discontinuation decisions were analyzed to determine whether they were patient-initiated or physician-initiated.
The study linked structured prescription records with unstructured clinical notes to ensure accurate identification of discontinuation.
Insights gained from this research can inform clinical decision-making and improve medication adherence strategies.
Prior studies have shown the potential of LLMs in extracting information from unstructured clinical data.
Clinical Implications
The findings highlight the importance of systematically capturing reasons for medication discontinuation in EHRs, which can enhance understanding of patient behavior and inform clinical practices. Utilizing LLMs may improve the analysis of clinical narratives, leading to better management of chronic diseases.
Conclusion
This study demonstrates the utility of LLMs in analyzing medication discontinuation, providing valuable insights for chronic disease management in Estonia. The approach may serve as a model for similar research in other contexts.
