RECOVER: Designing a Large Language Model-based Remote Patient Monitoring System for Postoperative Gastrointestinal Cancer Care

TL;DR

This paper presents RECOVER, an LLM-powered remote patient monitoring system designed for postoperative GI cancer care. It combines a conversational agent for patients with an interactive provider dashboard, embedding clinical guidelines and information needs into AI-driven workflows. Through participatory design sessions with patients and clinicians and two pilot studies, the work yields six design strategies, details system architecture, and demonstrates high usability and perceived value for RPM. The study also discusses responsible AI considerations, privacy, and deployment challenges, outlining path toward clinically integrated, scalable LLM-assisted RPM in high-risk surgical contexts.

Abstract

Cancer surgery is a key treatment for gastrointestinal (GI) cancers, a group of cancers that account for more than 35% of cancer-related deaths worldwide, but postoperative complications are unpredictable and can be life-threatening. In this paper, we investigate how recent advancements in large language models (LLMs) can benefit remote patient monitoring (RPM) systems through clinical integration by designing RECOVER, an LLM-powered RPM system for postoperative GI cancer care. To closely engage stakeholders in the design process, we first conducted seven participatory design sessions with five clinical staff and interviewed five cancer patients to derive six major design strategies for integrating clinical guidelines and information needs into LLM-based RPM systems. We then designed and implemented RECOVER, which features an LLM-powered conversational agent for cancer patients and an interactive dashboard for clinical staff to enable efficient postoperative RPM. Finally, we used RECOVER as a pilot system to assess the implementation of our design strategies with four clinical staff and five patients, providing design implications by identifying crucial design elements, offering insights on responsible AI, and outlining opportunities for future LLM-powered RPM systems.

Figures (12)

• Figure 1: PD sessions and their participants, discussion artifacts, and agenda.
• Figure 2: A PD session where participants commented on the conversation flow, and dashboard design version; the research team confirmed the participants' feedback by drawing lines and writing notes
• Figure 3: Dashboard Iteration Process. In each section, we present the key design versions of the corresponding module, together with the provider feedback that we gathered in the PD sessions. The provider's feedback guided us through the design iterations.
• Figure 4: System architecture of RECOVER. The red, purple, and blue arrows represent data generated by the Conversation Module, Information Extraction Module, and Summarization Module, respectively.
• Figure 5: Final Design of the RECOVER Dashboard. We present three key sections and the major interaction flow that connects them: (1) from patient list to patient detail, (2a) from patient list to key questions, (2b) from key questions visualization to detailed log, and (3) from daily report to summary. Each section also includes local interactions to review and manage patient reports.