NCT06689059

Research on Multimodal Multi-objective Integrated Machine Algorithm for Hip Replacement Surgery

HoPreM Platform: Efficient Multimodal Multi-Task Prediction of Perioperative Events Following Hip Replacement Surgery

• Status: Active Not Recruiting
• Phase: —
• Study type: Observational
• Enrollment: 6,271 (actual)
• Sponsor: Jingkun Liu · Academic / Other
• Sex: All
• Age: 18 Years
• Healthy volunteers: Not accepted

Summary

Purpose: The aim of this study is to develop the Holistic Predictive Multi-Tasking Platform for Clinical Data Analysis (HoPreM) to accurately predict perioperative events following hip replacement surgery by integrating various types of data, including demographic, surgical, medical history, and laboratory information. The events targeted for prediction include acute kidney injury (AKI), blood transfusion requirements, 48-hour postoperative discharge (48hPOD), Intensive Care Unit (ICU) transfer, and length of hospital stay (LOS). Key Questions: Can the HoPreM platform reduce the risk of complications after hip replacement surgery? How accurate is the platform in predicting the specified perioperative events? Participants: Participants will include patients undergoing hip replacement surgery, aged 18 and above, with less than 10% missing values in their medical records. The collected data will be used to train and test the predictive models of the HoPreM platform. Study Procedures: Patient data will be collected from Xi'an Honghui Hospital, including creatinine values recorded before and after surgery. The HoPreM platform will process multimodal data, including demographic, surgical, medical history, and laboratory test data. Various ensemble learning algorithms (including XGBoost, random forest, LightGBM, and CatBoost) will be applied to predict different perioperative outcomes. Expected Outcomes: The HoPreM platform is expected to demonstrate its capability in predicting complications after hip replacement surgery, particularly acute kidney injury and blood transfusion requirements. Through SHAP value analysis, the study aims to reveal relationships between features and clinical outcomes, enhancing the model's interpretability and clinical utility. Contact Information: For any questions about this study or for more information, please contact the research team.

Detailed description

This study aims to develop the Holistic Predictive Multi-Tasking Platform for Clinical Data Analysis (HoPreM) to accurately predict perioperative events following hip replacement surgery. The HoPreM platform integrates various types of patient data, including demographic, surgical, medical history, and laboratory information. Utilizing a multi-task learning framework, the platform is designed to predict multiple perioperative complications, such as acute kidney injury (AKI), blood transfusion requirements, 48-hour postoperative discharge (48hPOD), Intensive Care Unit (ICU) transfer, and length of hospital stay (LOS). To enhance predictive accuracy, feature selection techniques like Lasso regression and random forest models are employed, followed by ensemble learning algorithms, including CatBoost. This predictive platform is expected to support personalized postoperative management, reduce complication rates, and improve clinical outcomes for hip replacement patients.

Conditions

• Hip Replacement Surgery

Interventions

Type	Name	Description
OTHER	Multimodal Data Integration and Multi-Task Learning	This study utilizes a multimodal data integration and multi-task learning approach to predict perioperative events after hip replacement surgery. By combining various data types, including demographics, surgical details, medical history, and lab results, the model enhances prediction accuracy for outcomes like AKI, blood transfusion needs, and ICU transfers. The use of ensemble learning algorithms such as CatBoost optimizes the platform's performance, offering a unique method for clinical decision support.

Timeline

• Start date: 2024-10-24
• Primary completion: 2024-10-31
• Completion: 2025-12-31
• First posted: 2024-11-14
• Last updated: 2024-11-18