THE INTEGRATION OF GENERATIVE AI IN CLINICAL DECISION SUPPORT AND PATIENT CARE.

Abstract

The emergence of large language models (LLMs) and generative artificial intelligence (GenAI) has introduced a transformative shift in clinical informatics, moving beyond rule-based and predictive analytics toward systems capable of synthesizing unstructured clinical narratives, generating differential diagnoses, and producing patient-facing communication in natural language. This paper presents a comprehensive framework for integrating generative AI into clinical decision support systems (CDSS) and examines its downstream effects on patient care quality, clinician workload, and diagnostic accuracy. We propose a retrieval-augmented generation (RAG) architecture coupled with a fine-tuned transformer backbone, evaluated across a simulated multi-specialty dataset comprising 64,200 de-identified clinical encounters spanning primary care, emergency medicine, and chronic disease management. The proposed system, termed Clinical-RAG, achieves a diagnostic concordance rate of 91.4% with attending physician consensus, a documentation time reduction of 38.7%, and a clinician-reported usability score of 4.6 out of 5 on the System Usability Scale (SUS). Comparative analysis against standalone LLM baselines, traditional rule-based CDSS, and unaided clinician performance demonstrates statistically significant improvements in both diagnostic accuracy and care plan completeness. Safety evaluation reveals a hallucination rate of 2.1% on high-stakes recommendations, substantially mitigated through a verification layer grounded in structured electronic health record (EHR) data and current clinical guidelines. The findings support a cautious but optimistic pathway toward generative AI augmentation of clinical workflows, contingent on robust grounding mechanisms, human-in-the-loop oversight, and continuous post-deployment monitoring.