Application of Large Language Models Techniques to Post-ICU Syndrome Management in Critically Ill Patients: A Fully Longitudinal Mixed Study

The Affiliated Hospital Of Guizhou Medical University

Status: Invitation-only

Conditions: Post-Intensive Care Syndrome

Treatments: Behavioral: Health Promotion Model-Based Optimized Program; Behavioral: Routine Care; Behavioral: LLM-Enhanced Optimized Program

Study type: Interventional

Funder types: Other

Identifiers

NCT07141420

2024-171

Details and patient eligibility

About

The goal of this clinical trial is to evaluate whether Large Language Models (LLMs) combined with an optimized care program can effectively manage Post-Intensive Care Syndrome (PICS) in adult ICU survivors (aged ≥18 years) discharged from a tertiary hospital in China. The main questions it aims to answer are:

• Does the intervention (optimized program + LLMs) improve physical, psychological, cognitive, and social function recovery compared to standard care or the optimized program alone?
• How do patients experience and perceive the utility of LLMs in PICS self-management during recovery?

Researchers will compare three groups:

1. Group A (routine care)
2. Group B (optimized program without LLMs)
3. Group C (optimized program + LLMs) to see if adding LLMs significantly enhances PICS symptom management, patient self-efficacy, and quality of life over 6 months post-discharge.

Participants will:

• Install and use the Kimi Smart Assistant LLM (Group C only) for health queries under nurse supervision.

• Complete standardized questionnaires at discharge (baseline), 7 days, 1 month, 3 months, and 6 months post-discharge:

  • PICS Symptom Questionnaire (PICSQ)
  • Pittsburgh Sleep Quality Index (PSQI)
  • Anxiety (GAD-7) and Depression (PHQ-9) scales
  • Self-Management Ability Scale (AHSMSRS)

• Attend semi-structured interviews (Group C only) at 3 and 6 months to share experiences with LLM use.

Enrollment: 90 estimated patients

Sex: All

Ages: 18 to 100 years old

Volunteers: No Healthy Volunteers

Inclusion criteria

• ICU hospitalization duration > 24 hours.
• Age ≥ 18 years.
• Conscious at ICU discharge, able to communicate without barriers.
• Provide informed consent to participate.
• Regular access to and usage of smart electronic devices.

Exclusion criteria

• Previous ICU admission (≥24h) within 3 months before the current hospitalization.
• Transferred to another ICU during the current hospitalization.
• Pre-existing cognitive impairment (Blessed Dementia Rating Scale [BDRS] score >4 before ICU admission).
• Severe communication barriers:

Hearing impairment Dysarthria Other conditions preventing follow-up assessments.

• Critically unstable condition preventing questionnaire completion.
• Infrequent/no experience using smart electronic devices (e.g., smartphones, tablets).

Trial design

Primary purpose: Supportive Care

Allocation: Randomized

Interventional model: Parallel Assignment

Masking: Quadruple Blind

---

90 participants in 3 patient groups

Routine Care Group

Other group

Description:

Participants receive standard post-ICU follow-up care according to hospital protocols . This includes routine health assessments and general rehabilitation guidance at designated intervals (discharge, 7 days, 1/3/6 months post-discharge). No structured PICS management program or AI technology is provided.

Treatment:

Behavioral: Routine Care

Optimized Program Group

Other group

Description:

Participants receive an evidence-based, optimized PICS management program developed using the Health Promotion Model (HPM). This includes personalized rehabilitation plans, psychological support, and education tailored to PICS symptoms. Interventions are delivered by clinical staff at discharge, 7 days, and 1/3/6 months post-discharge. No AI/LLM technology is used.

Treatment:

Behavioral: LLM-Enhanced Optimized Program

Behavioral: Health Promotion Model-Based Optimized Program

Optimized Program + LLMs Group

Other group

Description:

Participants receive the same optimized PICS program as Group B, enhanced with Large Language Models (LLMs). Key components: Personalized AI-generated plans: ChatGPT-4 synthesizes patient data (baseline + follow-ups) to create monthly rehabilitation plans, reviewed by a multidisciplinary expert team. LLM access: Installation of "Kimi Smart Assistant" for daily health queries. Safety protocols: Patients must validate LLM advice with nurses via WeChat before use . Phased intervention: Pre-discharge: LLM training + baseline plan generation. 1 month: Plan updates based on new data. 3/6 months: Plan updates + semi-structured interviews about LLM experience.

Treatment:

Behavioral: LLM-Enhanced Optimized Program