AI-Powered Social Determinants of Health Extraction from Patient Records: A GPT-Based Investigation

Social Determinants of Health (SDOH) play a critical role in understanding the factors affecting patient well-being, and the identification of addiction status is a crucial component in this realm. This paper investigates the use of GPT (Generative Pretrained Transformer) model for extraction of addiction status from patient discharge summaries. The study demonstrates the capacity of GPT -based natural language processing to effectively parse patient notes and ascertain the presence or absence of addiction-related information. To harness the capability of this model and align it with our objective, prompt engineering was adopted with involving the iterative exploration of various prompt. The results reveal the model's remarkable capability to discern addiction-related information within patient records with a high accuracy. By effectively identifying addiction status, clinical care teams can address addiction-related issues promptly and comprehensively, leading to more targeted support and improved patient outcomes.