Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Retrieval-Augmented Framework for LLM-Based Clinical Decision Support

Leon Garza, Anantaa Kotal, Michael A. Grasso, Emre Umucu

TL;DR

The paper tackles the challenge of safe, context-aware prescribing in the face of heterogeneous EHR data by proposing a retrieval-augmented generation framework that grounds LLM outputs in semantically similar historical cases. It formalizes the problem as a retrieval-augmented conditional generation task, introduces a modular five-component architecture for fusing structured and unstructured EHR signals, and implements a retrieval and generation pipeline that pairs patient context with precedent cases to produce actionable treatment recommendations. The evaluation on a real ED dataset demonstrates competitive predictive performance relative to traditional baselines, with strong clinical consistency and interpretability due to precedent grounding. The work highlights practical considerations for deployment, including transparency, safety, and alignment with clinical workflows, and argues for the viability of LLM-assisted prescribing under appropriate validation and human-in-the-loop oversight.

Abstract

The increasing complexity of clinical decision-making, alongside the rapid expansion of electronic health records (EHR), presents both opportunities and challenges for delivering data-informed care. This paper proposes a clinical decision support system powered by Large Language Models (LLMs) to assist prescribing clinicians. The system generates therapeutic suggestions by analyzing historical EHR data, including patient demographics, presenting complaints, clinical symptoms, diagnostic information, and treatment histories. The framework integrates natural language processing with structured clinical inputs to produce contextually relevant recommendations. Rather than replacing clinician judgment, it is designed to augment decision-making by retrieving and synthesizing precedent cases with comparable characteristics, drawing on local datasets or federated sources where applicable. At its core, the system employs a retrieval-augmented generation (RAG) pipeline that harmonizes unstructured narratives and codified data to support LLM-based inference. We outline the system's technical components, including representation representation alignment and generation strategies. Preliminary evaluations, conducted with de-identified and synthetic clinical datasets, examine the clinical plausibility and consistency of the model's outputs. Early findings suggest that LLM-based tools may provide valuable decision support in prescribing workflows when appropriately constrained and rigorously validated. This work represents an initial step toward integration of generative AI into real-world clinical decision-making with an emphasis on transparency, safety, and alignment with established practices.

Retrieval-Augmented Framework for LLM-Based Clinical Decision Support

TL;DR

The paper tackles the challenge of safe, context-aware prescribing in the face of heterogeneous EHR data by proposing a retrieval-augmented generation framework that grounds LLM outputs in semantically similar historical cases. It formalizes the problem as a retrieval-augmented conditional generation task, introduces a modular five-component architecture for fusing structured and unstructured EHR signals, and implements a retrieval and generation pipeline that pairs patient context with precedent cases to produce actionable treatment recommendations. The evaluation on a real ED dataset demonstrates competitive predictive performance relative to traditional baselines, with strong clinical consistency and interpretability due to precedent grounding. The work highlights practical considerations for deployment, including transparency, safety, and alignment with clinical workflows, and argues for the viability of LLM-assisted prescribing under appropriate validation and human-in-the-loop oversight.

Abstract

The increasing complexity of clinical decision-making, alongside the rapid expansion of electronic health records (EHR), presents both opportunities and challenges for delivering data-informed care. This paper proposes a clinical decision support system powered by Large Language Models (LLMs) to assist prescribing clinicians. The system generates therapeutic suggestions by analyzing historical EHR data, including patient demographics, presenting complaints, clinical symptoms, diagnostic information, and treatment histories. The framework integrates natural language processing with structured clinical inputs to produce contextually relevant recommendations. Rather than replacing clinician judgment, it is designed to augment decision-making by retrieving and synthesizing precedent cases with comparable characteristics, drawing on local datasets or federated sources where applicable. At its core, the system employs a retrieval-augmented generation (RAG) pipeline that harmonizes unstructured narratives and codified data to support LLM-based inference. We outline the system's technical components, including representation representation alignment and generation strategies. Preliminary evaluations, conducted with de-identified and synthetic clinical datasets, examine the clinical plausibility and consistency of the model's outputs. Early findings suggest that LLM-based tools may provide valuable decision support in prescribing workflows when appropriately constrained and rigorously validated. This work represents an initial step toward integration of generative AI into real-world clinical decision-making with an emphasis on transparency, safety, and alignment with established practices.

Paper Structure

This paper contains 19 sections, 1 equation, 1 figure, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Task Definition
  4. System Architecture Overview
  5. Data Ingestion and Preprocessing
  6. Retrieval-Augmented Generation Framework
  7. RAG Inference Algorithm
  8. Evaluation
  9. Dataset
  10. Baseline Comparisons
  11. Evaluation Metrics
  12. Predictive Performance on Binary Classification Tasks
  13. Clinical Consistency of Recommendations
  14. Related Work
  15. Automated Clinical Decision Support Systems
  16. ...and 4 more sections

Figures (1)

  • Figure 1: Overview of the proposed prescribing support architecture. Structured and unstructured EHR data are preprocessed and encoded into embeddings. Relevant historical cases are retrieved using a similarity search, then combined with the current patient profile to form an augmented prompt for the language model. The LLM generates a ranked list of recommended treatments, optionally flagged for safety checks.