Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transforming Tuberculosis Care: Optimizing Large Language Models For Enhanced Clinician-Patient Communication

Daniil Filienko, Mahek Nizar, Javier Roberti, Denise Galdamez, Haroon Jakher, Sarah Iribarren, Weichao Yuwen, Martine De Cock

TL;DR

The paper tackles TB's global mortality burden by proposing a human-in-the-loop, LLM-powered digital adherence tool to enhance clinician-patient communication in resource-limited, multilingual settings. It combines prompt engineering, Retrieval-Augmented Generation (RAG), and privacy-preserving text sanitization to deliver empathetic yet medically accurate TB support in Argentinian Spanish. Through a multi-criteria evaluation (empathy, medical accuracy, linguistic relevance) across privacy configurations, the study reveals a trade-off between privacy and utility, with RAG sometimes reducing factual quality but enabling richer domain access. The work advances practical TB care by outlining a privacy-conscious, culturally attuned framework for deploying AI-assisted treatment support, while identifying key challenges and directions for future refinement and real-world validation.

Abstract

Tuberculosis (TB) is the leading cause of death from an infectious disease globally, with the highest burden in low- and middle-income countries. In these regions, limited healthcare access and high patient-to-provider ratios impede effective patient support, communication, and treatment completion. To bridge this gap, we propose integrating a specialized Large Language Model into an efficacious digital adherence technology to augment interactive communication with treatment supporters. This AI-powered approach, operating within a human-in-the-loop framework, aims to enhance patient engagement and improve TB treatment outcomes.

Transforming Tuberculosis Care: Optimizing Large Language Models For Enhanced Clinician-Patient Communication

TL;DR

The paper tackles TB's global mortality burden by proposing a human-in-the-loop, LLM-powered digital adherence tool to enhance clinician-patient communication in resource-limited, multilingual settings. It combines prompt engineering, Retrieval-Augmented Generation (RAG), and privacy-preserving text sanitization to deliver empathetic yet medically accurate TB support in Argentinian Spanish. Through a multi-criteria evaluation (empathy, medical accuracy, linguistic relevance) across privacy configurations, the study reveals a trade-off between privacy and utility, with RAG sometimes reducing factual quality but enabling richer domain access. The work advances practical TB care by outlining a privacy-conscious, culturally attuned framework for deploying AI-assisted treatment support, while identifying key challenges and directions for future refinement and real-world validation.

Abstract

Tuberculosis (TB) is the leading cause of death from an infectious disease globally, with the highest burden in low- and middle-income countries. In these regions, limited healthcare access and high patient-to-provider ratios impede effective patient support, communication, and treatment completion. To bridge this gap, we propose integrating a specialized Large Language Model into an efficacious digital adherence technology to augment interactive communication with treatment supporters. This AI-powered approach, operating within a human-in-the-loop framework, aims to enhance patient engagement and improve TB treatment outcomes.

Paper Structure

This paper contains 34 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Model development
  5. Prompt Engineering
  6. Privacy and Data Security
  7. Evaluation of models
  8. Results
  9. Empathy
  10. Medical Accuracy
  11. Linguistic Relevance
  12. Overall Quality
  13. Privacy
  14. Discussion
  15. Acknowledgments
  16. ...and 19 more sections

Figures (2)

  • Figure 1: The user's query will pass to the LLM-based AI system for processing. The clinical treatment supporter will receive top k suggested responses from the AI system and send the most fitting response to the patient.
  • Figure 2: The system classifies a patient's query as an "informational" or "emotional" request. Then, according to the classification result, an LLM is set up with the corresponding prompt and given access to external documents containing medical knowledge for informational questions.