The AIM-HEALTH study protocol, published in JMIR Research Protocols, introduces an AI-driven approach to improving patient comprehension of hospital discharge reports (HDRs). Researchers aim to enroll 200 adults from cardiology and nephrology units at the Azienda Socio-Sanitaria Territoriale Spedali Civili hospital. The study uses agentic AIs (autonomous systems capable of executing complex workflows) powered by open-weight large language models to generate two outputs: a patient-tailored supplementary discharge document (SDD) and a clinical informational performance report (CIPR) that checks for discrepancies between the discharge report and patient interviews.

Clinicians validate the AI-generated documents using a QUEST-informed tool, which evaluates accuracy, clarity, and safety. Only verified documents reach patients, and any errors serve as feedback to refine the system. Patient feedback will be gathered at a ~1-month follow-up to measure the tool's accessibility and usefulness. Data protection measures include on-premise processing and pseudonymization. The study, approved on February 24, 2026, began data collection on May 29, 2026, to assess technical feasibility and safety outcomes.

The AIM-HEALTH study protocol, published in JMIR Research Protocols, introduces an AI-driven approach to improving patient comprehension of hospital discharge reports (HDRs). Researchers aim to enroll 200 adults from cardiology and nephrology units at the Azienda Socio-Sanitaria Territoriale Spedali Civili hospital. The study uses agentic AIs (autonomous systems capable of executing complex workflows) powered by open-weight large language models to generate two outputs: a patient-tailored supplementary discharge document (SDD) and a clinical informational performance report (CIPR) that checks for discrepancies between the discharge report and patient interviews.

Clinicians validate the AI-generated documents using a QUEST-informed tool, which evaluates accuracy, clarity, and safety. Only verified documents reach patients, and any errors serve as feedback to refine the system. Patient feedback will be gathered at a ~1-month follow-up to measure the tool's accessibility and usefulness. Data protection measures include on-premise processing and pseudonymization. The study, approved on February 24, 2026, began data collection on May 29, 2026, to assess technical feasibility and safety outcomes.