Status
Conditions
About
Postoperative acute kidney injury (AKI) is a serious complication often linked to low blood pressure during surgery. This study aims to better protect patients' kidneys by personalizing how blood pressure is managed during an operation.
The project has two main goals:
First, researchers will analyze data from over 44,000 past surgeries to identify which specific blood pressure measurements are the most critical warning signs for kidney damage.
Second, using this knowledge, they will build a smart tool (a machine learning model) to predict a unique, safe blood pressure target for each individual patient before their surgery begins.
This personalized approach is intended to give doctors a specific target to maintain during surgery, helping to prevent kidney injury and improve patient safety.
Full description
Study Rationale Postoperative Acute Kidney Injury (AKI) is a common and severe complication following non-cardiac surgery, leading to prolonged hospital stays, increased medical costs, and a higher risk of persistent kidney failure or death. Intraoperative hypotension (abnormally low blood pressure during surgery) is recognized as a significant contributor to AKI, as it can reduce blood flow and oxygen supply to the kidneys.
Currently, there is no clinical consensus on which component of blood pressure-such as systolic (SBP), diastolic (DBP), or mean arterial pressure (MAP)-is the most critical to monitor for preventing organ injury. Furthermore, current guidelines often recommend a universal "one-size-fits-all" threshold for hypotension (e.g., MAP < 65 mmHg). This approach fails to account for individual patient differences, such as baseline blood pressure and co-existing health conditions, which may mean that the optimal blood pressure target varies significantly from person to person.
This study aims to address these gaps by using a large, multi-center dataset to first identify the most critical blood pressure components linked to AKI and then to develop a tool that predicts a personalized, optimal blood pressure threshold for individual patients.
Study Objectives
This study is divided into two parts:
Part 1: Risk Assessment and Blood Pressure Component Analysis
Part 2: Development of a Personalized Prediction Model
Primary Objective: To develop and validate a dynamic machine learning model that provides an individualized prediction of the optimal intraoperative hypotension threshold for patients undergoing non-cardiac surgery.
Secondary Objectives:
To build a machine learning model that predicts a patient's risk of developing postoperative AKI based on their preoperative data.
②To explore the relationship between the cumulative 5-minute lowest intraoperative blood pressure and postoperative AKI, using this relationship to refine the predictive model.
3. Study Design and Methods This is a retrospective, multi-center cohort study.
Data Sources: The study will utilize de-identified electronic health record data from several large sources: the public INSPIRE database, Zhongda Hospital affiliated with Southeast University (2013-2024), Nanjing First Hospital (2016-2024), and the First Affiliated Hospital of Anhui Medical University (2013-2023).
Study Population: The study will include adult patients (age ≥ 18) who underwent non-cardiac surgery and had available intraoperative blood pressure monitoring data. An initial dataset of approximately 850,000 surgical records will be screened, with an estimated final cohort of 44,504 eligible patients after applying inclusion and exclusion criteria. Key exclusions include patients with pre-existing severe chronic kidney disease, those undergoing urological or kidney-related surgeries, and procedures with a short duration of anesthesia (<60 min).
In Part 1, the research team will use unsupervised clustering methods (e.g., K-means) to group patients based on preoperative characteristics. Subsequently, multivariable logistic regression models will be used to analyze the association between intraoperative hypotension metrics and the risk of postoperative AKI/AKD within the overall cohort and across the different patient clusters.
In Part 2, machine learning algorithms, including XGBoost and Random Forest, will be employed to develop a series of predictive models. The final model will be designed to dynamically adjust a patient's hypotension threshold by iteratively calculating the AKI risk until it falls below a predefined safety level (10%). The model's performance will be rigorously evaluated using metrics such as the Area Under the Curve (AUC), and its interpretability will be assessed using SHAP (Shapley Additive Explanations) analysis.
Enrollment
Sex
Ages
Volunteers
Inclusion criteria
Exclusion criteria
Loading...
Central trial contact
Ran You Wang
Data sourced from clinicaltrials.gov
Clinical trials
Research sites
Resources
Legal