This study is a multicenter, prospective, randomized controlled trial designed to evaluate the effectiveness and safety of a multimodal artificial intelligence (AI)-guided postoperative recovery management system in patients after lung cancer surgery. Eligible patients will be enrolled after surgery when their clinical condition is stable and will be randomly assigned to either an AI-guided recovery management group or a usual postoperative care group.
Patients in the AI-guided group will receive usual postoperative care plus a multimodal AI-based recovery management system. The system will collect patient-reported symptoms, vital signs, physical activity, respiratory rehabilitation information, recovery-related data, and, when needed, wound or chest-related images or short videos. Based on these data, the system will provide recovery feedback, general nursing advice, respiratory rehabilitation reminders, activity guidance, and risk stratification alerts. For red-flag symptoms or high-risk conditions, the system will advise patients to contact the clinical team or seek medical care.
Patients in the usual-care group will receive standard postoperative management after lung cancer surgery and will complete symptom assessments at the same prespecified time points, but they will not receive AI-generated individualized recovery feedback or AI-generated risk alerts.
The primary outcome is the number of MDASI-LC-derived target symptom threshold events within 30 days after surgery. Target symptoms include pain, fatigue, disturbed sleep, shortness of breath, and cough. Secondary outcomes include overall target symptom burden, quality of recovery, time to recovery to a mild-symptom state, functional interference, respiratory rehabilitation adherence, physical activity adherence, unplanned healthcare utilization, pulmonary complications, and unplanned readmission.
Age 18 years or older.
Clinically or pathologically diagnosed with lung cancer.
Undergoing lung cancer-related thoracic surgery.
Surgical procedures may include video-assisted thoracoscopic surgery, robot-assisted thoracic surgery, or open thoracic surgery.
Patients receiving non-surgical treatment only.
Patients undergoing bronchoscopy, percutaneous biopsy, or other non-surgical diagnostic or therapeutic procedures only.
Clinically unstable after surgery and unable to participate in symptom assessment or recovery management.
Requiring prolonged intensive care unit treatment, continuous advanced life support, or continuous intensive medical management.
Parallel
Conventional Internet-Based Rehabilitation Information
BehavioralParticipants in the conventional internet group will receive routine postoperative rehabilitation nursing and use conventional internet-based information resources for rehabilitation-related information after thoracic surgery.
Large Language Model-Assisted Rehabilitation Nursing
BehavioralParticipants in the large language model-assisted rehabilitation nursing group will receive artificial intelligence-assisted postoperative rehabilitation nursing. The system will analyze postoperative patient data and generate individualized rehabilitation recommendations, including pain management, pulmonary function training, and exercise rehabilitation. The rehabilitation module will support digital entry and feedback of standardized scales and recovery-related information, including pain scores, activity monitoring, step count, symptom self-assessment, cough frequency, and dyspnea level. Healthcare professionals will remain responsible for clinical oversight and safety management.
Number of MDASI-LC-Derived Target Symptom Threshold Events Within 30 Days After Surgery
The outcome is the number of target symptom threshold events per patient from postoperative intervention initiation to postoperative day 30. Target symptoms include pain, fatigue, disturbed sleep, shortness of breath, and cough. Each target symptom will be assessed using MDASI-LC items or equivalent 0-10 numeric rating scales. A target symptom threshold event is defined as any target symptom score of 4 or higher at a prespecified assessment time point.
Overall Target Symptom Burden Within 30 Days After Surgery
Overall target symptom burden will be assessed using MDASI-LC target symptom items, including pain, fatigue, disturbed sleep, shortness of breath, and cough. The outcome may be summarized as the mean score of the five target symptoms at each prespecified time point or as the area under the symptom burden curve over 30 days. Higher scores indicate greater symptom burden.
Postoperative quality of recovery score
Postoperative quality of recovery will be assessed using the QoR-15 or QoR-40 questionnaire. The questionnaire evaluates domains such as pain, physical comfort, emotional state, physical independence, and psychological support. Higher scores indicate better postoperative recovery quality.
Time to Recovery to a Mild-Symptom State
Time to recovery to a mild-symptom state is defined as the time from postoperative intervention initiation to the first day when all target symptom scores are below 4 and remain below 4 for at least 3 consecutive days. Target symptoms include pain, fatigue, disturbed sleep, shortness of breath, and cough.
MDASI-LC Functional Interference Score
Functional interference will be assessed using MDASI-LC interference items. These items evaluate the extent to which postoperative symptoms interfere with daily function, including general activity, walking, work or household tasks, mood, relations with others, and enjoyment of life. Higher scores indicate greater functional interference.
AI High-Risk Alert Performance
This outcome evaluates the performance of AI-generated high-risk alerts in the AI-guided recovery management group. It includes the incidence of AI-generated high-risk alerts and the missed high-risk alert rate. The incidence is defined as the proportion of participants who receive at least one AI-generated high-risk alert during the postoperative monitoring period. The missed high-risk alert rate is defined as the proportion of participants with clinically confirmed postoperative complications, severe symptom deterioration, or conditions requiring escalation of care who did not receive an AI-generated high-risk alert within the prespecified monitoring window.
Incidence of Clinically Inappropriate AI-Generated Recommendations
This outcome evaluates the proportion of AI-generated recommendations judged by an independent clinical safety review committee to be clinically inappropriate, unsafe, potentially misleading, or inconsistent with prespecified postoperative recovery management rules.
AI Monitoring Feasibility and Data Quality
This outcome evaluates the feasibility of AI-based recovery monitoring and the adequacy of uploaded data in the AI-guided recovery management group. It includes patient adherence to AI-based recovery monitoring and the data quality adequacy rate. Patient adherence is defined as the proportion of prespecified monitoring days on which participants complete symptom reporting, vital sign upload, rehabilitation behavior recording, and required image or video upload, where applicable. Data quality adequacy is defined as the proportion of uploaded symptom data, vital sign data, rehabilitation records, and image or video data that meet prespecified quality criteria, including completeness, clarity, interpretability, and timestamp completeness.
Click on the lead sponsor to view all of their trials.
