Derivation and Verification of Hemodynamic Clinical Subphenotypes in Patients Undergoing Cardiac Surgery Under Unsupervised Machine Learning

Background & Objective:

Cardiac surgery patients differ significantly in their health conditions and how they react during operations. Standard risk assessments before surgery often miss the real-time changes happening inside a patient's body during the procedure, which can affect their recovery. Therefore, researchers conducted this study to find different groups (phenotypes) of patients who face varying risks for poor outcomes. They did this by using advanced computer learning techniques to analyze a lot of detailed health information collected both before and during surgery.

Methods:

This was a study that looked back at patient records from several hospitals. Researchers gathered a large amount of patient information from before surgery, including their basic health details and lab results. They also collected very detailed measurements of patients' vital signs taken during surgery, noting how these changed over time. Then, a computer program that can find patterns without being told what to look for (unsupervised hierarchical clustering) was used to sort patients into distinct groups based on this combined data.

Clinical Relevance:

This study expects to show that using data to identify patient groups can reveal differences that traditional methods miss. These new patient groups, which are based on how their blood flow and vital signs behave, offer a new way to understand risks in real-time. This could help doctors to predict problems more accurately and create personalized care plans for each patient around the time of surgery, which has great potential for practical use in hospitals.