A Predictive Model for Postoperative Delirium in Kidney Transplant Patients

Construction of a Machine Learning Prediction Model for Postoperative Delirium in Kidney Transplant Patients Based on Clinical Data

This study aims to develop and prospectively validate a machine learning-based prediction model for postoperative delirium in kidney transplant recipients, using perioperative clinical data. Delirium is a common and serious postoperative complication that significantly increases morbidity, mortality, and healthcare costs. By analyzing electronic medical records from kidney transplant patients, including preoperative, intraoperative, and postoperative variables, the study seeks to identify high-risk patients and key predictors. Six machine learning models, including XGBoost, LGBM, GBC, LR, ANN, and SVM, will be constructed and evaluated, with a soft voting ensemble classifier used to optimize prediction performance. The goal is to improve early recognition and clinical management of postoperative delirium in kidney transplant patients.

Inclusion Criteria:

• Patients who underwent kidney transplantation at [Hospital Name, if needed] between January 1, 2016, and May 31, 2025.

Age ≥ 18 years at the time of transplantation.

Discharged alive from the hospital after surgery.

Complete perioperative clinical data available, including preoperative evaluations, intraoperative records, and postoperative documentation

Exclusion Criteria:

• Patients with documented pre-existing delirium or major neurocognitive disorder before surgery.

Simultaneous or multi-organ transplantation (e.g., kidney-pancreas).

Death within 7 days postoperatively.

Incomplete or missing key electronic medical records preventing outcome assessment.

Patients who withdrew consent for use of clinical data for research purposes (for prospective part).