Academic

Automated Auditing of Hospital Discharge Summaries for Care Transitions

arXiv:2604.05435v1 Announce Type: new Abstract: Incomplete or inconsistent discharge documentation is a primary driver of care fragmentation and avoidable readmissions. Despite its critical role in patient safety, auditing discharge summaries relies heavily on manual review and is difficult to scale. We propose an automated framework for large-scale auditing of discharge summaries using locally deployed Large Language Models (LLMs). Our approach operationalizes core transition-of-care requirements such as follow-up instructions, medication history and changes, patient information and clinical course, etc. into a structured validation checklist of questions based on DISCHARGED framework. Using adult inpatient summaries from the MIMIC-IV database, we utilize a privacy-preserving LLM to identify the presence, absence, or ambiguity of key documentation elements. This work demonstrates the feasibility of scalable, automated clinical auditing and provides a foundation for systematic quality

A
Akshat Dasula, Prasanna Desikan, Jaideep Srivastava · · 1 min read · 3 views
#cs.AI

arXiv:2604.05435v1 Announce Type: new Abstract: Incomplete or inconsistent discharge documentation is a primary driver of care fragmentation and avoidable readmissions. Despite its critical role in patient safety, auditing discharge summaries relies heavily on manual review and is difficult to scale. We propose an automated framework for large-scale auditing of discharge summaries using locally deployed Large Language Models (LLMs). Our approach operationalizes core transition-of-care requirements such as follow-up instructions, medication history and changes, patient information and clinical course, etc. into a structured validation checklist of questions based on DISCHARGED framework. Using adult inpatient summaries from the MIMIC-IV database, we utilize a privacy-preserving LLM to identify the presence, absence, or ambiguity of key documentation elements. This work demonstrates the feasibility of scalable, automated clinical auditing and provides a foundation for systematic quality improvement in electronic health record documentation.

Executive Summary

The article presents an innovative framework for automating the auditing of hospital discharge summaries to address care fragmentation and avoidable readmissions. Leveraging locally deployed Large Language Models (LLMs) and the DISCHARGED framework, the study operationalizes transition-of-care requirements into a structured validation checklist. Using the MIMIC-IV database, the authors demonstrate the feasibility of a privacy-preserving LLM to systematically identify the presence, absence, or ambiguity of critical documentation elements. This approach aims to scale clinical auditing, enhance patient safety, and support quality improvement in electronic health record (EHR) documentation. The research underscores the potential of AI-driven solutions to transform healthcare documentation practices.

Key Points

  • ▸ The study addresses a critical gap in healthcare documentation by proposing an automated framework for auditing discharge summaries, which are prone to incompleteness or inconsistency, leading to care fragmentation and readmissions.
  • ▸ The framework utilizes locally deployed LLMs to operationalize core transition-of-care requirements into a structured validation checklist based on the DISCHARGED framework, enabling systematic and scalable auditing.
  • ▸ The research demonstrates feasibility using the MIMIC-IV database, showcasing the ability of privacy-preserving LLMs to identify key documentation elements and their ambiguities, thereby supporting patient safety and quality improvement initiatives.

Merits

Methodological Innovation

The study introduces a novel approach by combining LLMs with structured clinical frameworks (e.g., DISCHARGED) to automate auditing of discharge summaries, addressing a longstanding challenge in healthcare documentation.

Scalability and Efficiency

The framework’s reliance on LLMs enables large-scale, automated auditing, significantly reducing the manual burden and improving scalability compared to traditional review methods.

Privacy Preservation

The use of locally deployed LLMs ensures compliance with privacy regulations (e.g., HIPAA), mitigating risks associated with data sharing and enhancing trust in AI-driven healthcare solutions.

Clinical Relevance

By focusing on transition-of-care requirements, the framework directly addresses patient safety concerns, such as avoidable readmissions and care fragmentation, aligning with healthcare quality improvement goals.

Demerits

Dependence on Data Quality

The framework’s accuracy is inherently tied to the quality of the input data (e.g., discharge summaries in the MIMIC-IV database). Poorly documented or ambiguous records may lead to suboptimal performance or misclassification.

Generalizability Concerns

The study’s reliance on a single database (MIMIC-IV) may limit the generalizability of the results to other healthcare systems or regions with differing documentation practices or clinical workflows.

LLM Limitations

While LLMs offer significant advancements, they are not infallible. Potential issues such as hallucinations, bias, or misinterpretation of clinical nuances could undermine the reliability of the auditing framework.

Implementation Challenges

Deploying locally deployed LLMs in clinical settings requires robust infrastructure, technical expertise, and ongoing maintenance, which may pose barriers to adoption for smaller or resource-constrained healthcare facilities.

Expert Commentary

The article presents a compelling and timely contribution to the intersection of AI and healthcare, particularly in addressing the longstanding challenges of incomplete or inconsistent discharge documentation. By operationalizing the DISCHARGED framework into a structured validation checklist and leveraging locally deployed LLMs, the authors have demonstrated a scalable and privacy-preserving approach to auditing discharge summaries. This work is significant for several reasons. First, it addresses a critical patient safety issue by targeting care fragmentation and avoidable readmissions, which are often linked to poor documentation. Second, the methodological innovation of combining LLMs with clinical frameworks offers a replicable model for other healthcare documentation challenges. However, the study also raises important considerations. The dependence on high-quality input data and the potential for LLM limitations, such as hallucinations or bias, underscore the need for rigorous validation and continuous monitoring. Additionally, the framework’s reliance on a single dataset (MIMIC-IV) may limit its generalizability, necessitating further research across diverse healthcare systems. Overall, the article is a valuable addition to the literature, providing a foundation for future advancements in AI-driven clinical auditing while highlighting the importance of addressing the practical and ethical challenges associated with such technologies.

Recommendations

  • ✓ Conduct further validation studies across diverse healthcare systems and datasets to assess the framework’s generalizability and robustness.
  • ✓ Develop standardized protocols for input data quality and LLM deployment to mitigate risks such as hallucinations, bias, and misclassification.
  • ✓ Establish interdisciplinary collaboration between AI researchers, clinicians, and policymakers to address ethical, regulatory, and implementation challenges associated with AI-driven auditing frameworks.
  • ✓ Integrate the framework into broader healthcare quality improvement initiatives, such as value-based care programs, to maximize its impact on patient safety and outcomes.
  • ✓ Explore the potential for integrating the framework with other AI-driven tools, such as clinical decision support systems, to enhance its utility and comprehensiveness.

Sources

Original: arXiv - cs.AI

Related Articles