Knowledge-Guided Large Language Model for Automatic Pediatric Dental Record Understanding and Safe Antibiotic Recommendation

Summary

This study introduces a Knowledge-Guided Large Language Model (KG-LLM) designed for automated understanding of pediatric dental records and safe antibiotic recommendation. By integrating a specialized knowledge graph, retrieval-augmented generation (RAG), and a multi-stage safety validation pipeline, the system significantly improves record interpretation, antibiotic prescription accuracy, and notably reduces unsafe antibiotic suggestions compared to a baseline LLM.

Medical Relevance

This research is highly relevant to medicine and health by providing an advanced AI solution to enhance patient safety and treatment efficacy in pediatric dentistry. It directly addresses the problem of medication errors, particularly with antibiotics, in a vulnerable patient population, by ensuring evidence-grounded and safety-validated recommendations.

AI Health Application

The paper develops a Knowledge-Guided Large Language Model (KG-LLM) that utilizes a pediatric dental knowledge graph, retrieval-augmented generation (RAG), and a multi-stage safety validation pipeline. This AI system automatically interprets unstructured pediatric dental clinical records and radiology reports, extracts clinical entities, summarizes diagnoses, and provides evidence-grounded and safe antibiotic recommendations (drug, dose, duration). It also features a dual-layer validation mechanism to prevent unsafe suggestions related to allergies, contraindications, and dosing errors, thereby acting as an intelligent clinical decision support tool for safe medication prescription in dentistry.

Key Points

• Addresses the critical challenges of interpreting unstructured pediatric dental records and ensuring safe, accurate antibiotic prescribing.
• Proposes a KG-LLM framework that integrates a pediatric dental knowledge graph, retrieval-augmented generation (RAG), and a multi-stage safety validation pipeline.
• Utilizes a clinical NER/RE module to extract structured entities and relations from dental notes and radiology reports as input.
• Leverages the knowledge graph to retrieve relevant guidelines, drug-safety rules, and analogous historical cases to inform the LLM's diagnostic summarization and prediction.
• Incorporates a dual-layer safety validation mechanism, combining deterministic rule checking with a learned classifier, to detect allergies, contraindications, and dosing errors.
• Demonstrated substantial improvements on 32,000 de-identified pediatric dental visit records, achieving higher F1-score (0.914 vs. 0.867) for record understanding and Top-1 accuracy (0.782 vs. 0.716) for drug-dose-duration prediction against a Llama-2 clinical baseline.
• Crucially, the KG-LLM reduced unsafe antibiotic suggestions by 50%, highlighting a significant advancement in patient safety for pediatric dental care.

Clinical Impact

This KG-LLM has the potential to significantly improve the accuracy and safety of antibiotic prescribing in pediatric dental settings. By automating evidence-grounded recommendations and implementing robust safety checks, it can reduce medication errors, minimize adverse drug reactions, and enhance the quality of patient care. It offers a powerful clinical decision support tool that can assist dentists in navigating complex patient data and clinical guidelines, ultimately leading to better patient outcomes and optimized antibiotic stewardship.

Limitations

The abstract does not explicitly detail specific limitations of the proposed system or its evaluation. Potential implicit limitations (not stated) could include the generalizability of the knowledge graph to diverse patient populations or other dental specialties, or the ongoing maintenance burden of the knowledge graph and safety rules.

Future Directions

The abstract does not explicitly mention future research directions.

Medical Domains

Keywords

Abstract

Accurate interpretation of pediatric dental clinical records and safe antibiotic prescribing remain persistent challenges in dental informatics. Traditional rule-based clinical decision support systems struggle with unstructured dental narratives, incomplete radiographic descriptions, and complex safety constraints. To address these limitations, this study proposes a Knowledge-Guided Large Language Model (KG-LLM) that integrates a pediatric dental knowledge graph, retrieval-augmented generation (RAG), and a multi-stage safety validation pipeline for evidence-grounded antibiotic recommendation. The framework first employs a clinical NER/RE module to extract structured entities and relations from dental notes and radiology reports. Relevant guidelines, drug-safety rules, and analogous historical cases are subsequently retrieved from the knowledge graph and supplied to the LLM for diagnostic summarization and dose-drug-duration prediction. Safety assurance is achieved through a dual-layer validation mechanism combining deterministic rule checking with a learned classifier for detecting allergies, contraindications, and dosing errors. Experiments on 32,000 de-identified pediatric dental visit records demonstrate the effectiveness of the proposed approach. Compared with a domain-adapted Llama-2 clinical baseline, KG-LLM improves record-understanding performance (F1: 0.914 vs. 0.867), drug-dose-duration accuracy (Top-1: 0.782 vs. 0.716), and reduces unsafe antibiotic suggestions by 50%. Additional evaluation across summary quality, recommendation accuracy, and global safety scores further confirms the robustness of the system. Ablation analyses indicate that the knowledge graph, RAG, and safety modules each contribute substantially to clinical reliability and interpretability.