From Patient Consultations to Graphs: Leveraging LLMs for Patient Journey Knowledge Graph Construction

Hassan S. Al Khatib 1 , Sudip Mittal 1 , Shahram Rahimi 2 , Nina Marhamati 3 , Sean Bozorgzad 3 · 2025

Read From Patient Consultations to Graphs: Leveraging LLMs for Patient Journey Knowledge Graph Construction by Hassan S. Al Khatib 1 , Sudip Mittal 1 , Shahram Rahimi 2 , Nina Marhamati 3 , Sean Bozorgzad 3 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16533.txt.

Introduction

Hassan Al Khatib, Sudip Mittal, Shahram Rahimi, Nina Marhamati, and Sean Bozorgzad turn patient talks into knowledge graphs of the care journey. Their paper from twenty twenty-five tackles scattered health records. Care data sit in split systems with no whole view of a patient's path. Their graphs fuse formal notes with raw patient-provider chats into one structured map.

Large language models do the reading and structuring. Time and cause links join visits, diagnoses, treatments, and outcomes. Four models were graded on graph quality and speed. All four met the structural rules in full. They split on medical entity handling and compute cost. The authors map open hurdles and next steps. The aim is care that is planned around the person, with outcome forecasts in reach.

Catalog record

CREATORHassan S. Al Khatib 1 , Sudip Mittal 1 , Shahram Rahimi 2 , Nina Marhamati 3 , Sean Bozorgzad 3
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16533

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

From Patient Consultations to Graphs: Leveraging LLMs for Patient Journey Knowledge Graph Construction Hassan S. Al Khatib 1 , Sudip Mittal 1 , Shahram Rahimi 2 , Nina Marhamati 3 , Sean Bozorgzad 3 Abstract The transition towards patient-centric healthcare necessitates a comprehensive understanding of patient journeys, which encompass all healthcare experiences and interactions across the care spectrum. Existing healthcare data systems are often fragmented and lack a holistic representation of patient trajectories, creating challenges for coordinated care and personalized interventions. Patient Journey Knowledge Graphs (PJKGs) represent a novel approach to addressing the challenge of fragmented healthcare data by integrating diverse patient information into a unified, structured representation. This paper presents a methodology for constructing PJKGs using Large Language Models (LLMs) to process and structure both formal clinical documentation and unstructured patient-provider conversations. These graphs encapsulate temporal and causal relationships among clinical encounters, diagnoses, treatments, and outcomes, enabling advanced temporal reasoning and personalized care insights. The research evaluates four different LLMs, such as Claude 3.5, Mistral, Llama 3.1, and Chatgpt4o, in their ability to generate accurate and computationally efficient knowledge graphs. Results demonstrate that while all models achieved perfect structural compliance, they exhibited variations in medical entity processing and computational efficiency. The paper concludes by identifying key challenges and future research directions. This work contributes to advancing patient-centric healthcare through the development of comprehensive, actionable knowledge graphs that support improved care coordination and outcome prediction. I Introduction Understanding and optimizing a patient’s healthcare journey is essential to transitioning toward patient-centric healthcare. A patient journey encompasses all healthcare experiences and interactions, from the initial symptoms or diagnosis through treatment and follow-up care [ 1 ] . Despite its significance, patient data is often fragmented across multiple platforms and lacks a long-term, comprehensive representation [ 2 ] . This fragmentation creates significant challenges for healthcare providers in delivering coordinated care and for patients in effectively managing their health [ 3 ] . This limitation emphasizes the need for advanced frameworks to consolidate and structure disparate data sources into meaningful, actionable insights. In this context, PJKGs emerge as an innovative solution, offering a unified and holistic representation of patient encounters across the healthcare spectrum. Unlike traditional data models that emphasize provider-centric perspectives (such as episode-based care records and discrete clinical documentation), PJKGs prioritize patient-centeredness by integrating diverse elements such as clinical records, lifestyle factors, and social determinants of health. This integration enables pattern identification and outcome predictions. Furthermore, it facilitates more effective intervention personalization. This approach ensures a more comprehensive understanding of the factors influencing health outcomes. The motivation for developing PJKGs stems from several pressing challenges in healthcare data management and utilization. First, existing healthcare data models often operate in silos, making it difficult to trace a patient’s complete medical history across different providers and systems [ 5 ] . Second, traditional models often fail to account for the temporal, causal, and sequential relationships important for understanding patient trajectories. Third, integrating unconventional data sources, such as patient-provider encounter conversations, remains largely unexplored, leaving significant gaps in delivering genuinely personalized care. This research addresses these challenges by leveraging a healthcare journey ontology [ 19 ] while creating robust methods for integrating structured and unstructured healthcare data and establishing frameworks for temporal and causal relationship extraction in patient trajectories. Our research leverages and evaluates different LLMs to achieve these objectives in constructing PJKGs. LLMs offer unique capabilities in processing and understanding unstructured medical text, enabling the extraction of complex relationships and medical concepts from diverse data sources [ 6 ] . These knowledge graphs (KGs) are designed to map the temporal and causal relationships between clinical encounters, diagnoses, treatments, and outcomes, providing a dynamic and interconnected view of patient health trajectories. Our approach contributes to the field by introducing a novel framework for automated PJKG construction using LLMs that allows for temporal reasoning in healthcare journey mapping by integrating unstructured clinical conversations into structured knowledge graphs. Building upon these foundations and methodological innovations, we present the organization of our research as follows: Section 2 overviews existing healthcare KGs, focusing on the methodologies and applications highlighting various patient-centric approaches. Section 3 describes the proposed methodology using different LLMs, including ontology structure, data sources, and the PJKG building process. Section 4 discusses the evaluation of PJKGs through quantitative analyses. Section 5 presents a detailed discussion of the findings. Finally, Section 6 concludes the paper by summarizing the findings and outlining directions for future research. II Background and Literature Review This section explores how healthcare KGs have evolved from disease-centric models to patient-centric knowledge graphs (PCKGs), enabling more individualized and data-driven care. We also highlight PCKGs’ transformative potential in personalized treatment, clinical trial optimization, and disease prediction. II-A Advancements in Healthcare KGs The evolution of KGs in healthcare signifies a paradigm shift from traditional data systems to more complex and interconnected networks. Early implementations of KGs, such as those rooted in electronic health records (EHRs), primarily focused on disease-centric models that integrated multimodal data for clinical problem-solving [ 7 , 8 ] . These graphs helped explore relationships between diseases, symptoms, and treatments, though they often lacked the granularity required for patient-centric care [ 9 ] . Recent advancements have introduced PCKGs [ 4 ] , prioritizing the holistic integration of an individual’s medical history, genetic information, lifestyle behaviors, and environmental influences. PCKGs leverage structured and unstructured data sources, including clinical notes and diagnostic results, to create comprehensive patient profiles. The adoption of advanced machine learning techniques, such as named entity recognition (NER) and relationship extraction (RE), has enhanced the ability of KGs to map complex relationships within healthcare data [ 10 , 11 ] . PCKGs emphasize the temporal and causal relationships inherent in patients’ data, enabling dynamic representation of health trajectories. This evolution, supported by reasoning mechanisms and semantic search techniques, allows healthcare providers to derive actionable insights for clinical decision-making [ 12 ] . The emphasis on patient-specific data marks a critical advancement in KG technology, aligning with the goals of personalized and precision medicine. II-B Applications of PCKGs PCKGs will transform healthcare by facilitating various applications that enhance patient care, improve diagnostic accuracy, and enable predictive modeling. One of the most significant applications is the ability to personalize treatment plans by integrating data from diverse sources to tailor interventions for individual patients [ 13 ] . PCKGs have been used to develop personalized treatment regimens and predict adverse drug reactions based on a patient’s medical history and genetic profile. By analyzing relationships between genes, diseases, and treatments, PCKGs facilitate the identification of novel therapeutic targets and treatment approaches [ 15 ] . Another application is disease prediction, where PCKGs leverage advanced analytics to identify potential health risks before symptoms manifest. Methods such as graph neural networks (GNNs) and multi-hop reasoning enable the identification of complex patterns that traditional models might overlook [ 14 ] . These predictive capabilities support early intervention strategies and improve long-term health outcomes by proactively addressing conditions. PCKGs also play an essential role in clinical trials. They streamline patient recruitment by matching individuals to trials based on detailed eligibility criteria derived from their health data. This application has been shown to increase trial efficiency and ensure better outcomes by

Read the complete text