LLMs for Clinical-Research Data Extraction
Acute heart failure including cardiogenic shock is a life-threatening condition with high 30-day mortality up to 60%. In order to understand these critical conditions better, large registries are being established. These are most valuable primarily in generating hypotheses for further assessment, usually performed in randomized controlled trials (RCTs). These trials permit insights into causal relationships between medical interventions (like the use of novel medical drugs or mechanical circulatory support devices like the veno-arterial extracorporeal membrane oxygenation, so-called VA-ECMO) and patient outcomes.
Unfortunately, building the registries as well as performing RCTs are labor-intensive endeavors and, thus, both time-consuming and costly. Leveraging existing data collected from clinical routine is of utmost importance to advance the research in understanding critical conditions. Increasing the rate of patient recruitment by better embedding research-related tasks into clinical routine will reduce RCT durations necessary to obtain sufficient numbers of patients.
Applying artificial intelligence (AI) through large language models (LLMs) addresses this need: Much information of interest to the clinical researcher is contained in discharge letters from hospitals in a more or less structured way. Instead of spending labor force to collect these data, applying AI is a valuable and cost-saving alternative.
Objectives
- Automate data extraction from clinical discharge letters using LLMs.
- Support the creation of large registries for critical conditions like acute heart failure.
- Reduce the time and cost associated with conducting randomized controlled trials (RCTs).
- Improve patient recruitment by integrating research tasks into the clinical routine.