Artificial Intelligence - to Predict and Prevent Hypotension During Surgery

Overview

The goal of this medtech clinical trial is to develop and evaluate a machine learning algoritm to predict low blood pressure episodes during major surgery. The main questions it aims to answer are: * Could a novel method for cardiac output estimation through alterations in carbon dioxide improve the performance of a blood pressure based algoritm in order to predict low blood pressure episodes during major abdominal surgery? * Will the predictive performance of the algoritm improve with the addition of other patient specific data? * Do the estimated cardiac output and central venous saturation by the novel method agree with our invasive arterial pressure method for cardiac output, and samples via a central venous line, respectively? 300 participants will be anesthetized with total intravenous anesthesia and ventilated with the novel carbon dioxide based method, and arterial and central venous blood gases will be taken regularly throughout the operation. All physiological data will be stored for later analyses and development of the algoritm by machine learning methods. No other invasive interventions will be performed outside our standard clinical peroperative protocol.

Every year, approximately 800,000 patients are operated on in Sweden. Of them, a relatively high proportion suffer from serious complications after surgery, such as heart and kidney damage, stroke and even death. There is a demonstrated connection between the above complications and blood pressure drops during surgery, which means that in any case some of the complications are potentially avoidable if the blood pressure is kept at a stable and adequate level based on each individual patient's conditions. Unfortunately, episodes of a drop in blood pressure are difficult to predict with the standard monitoring methods available and, therefore unfortunately, often occur during surgery. At Karolinska University Hospital, we are now building a new system that, with the help of artificial intelligence, can sound the alarm even before the drop in blood pressure occurs. In the study, we collect and combine data from the standard monitoring during surgery; pulse, blood pressure, oxygenation, cardiac output, ECG, lab values from blood gases on patients undergoing abdominal surgery which is a group of patients who have a higher risk of suffering from postoperative complications. To this information, we will also add relevant data from a novel method for circulation monitoring based on variations in exhaled carbon dioxide. We will also record the reason why blood pressure has dropped, such as bleeding, drug impact, dehydration, with the aim that the AI algorithms should be able to distinguish between different causes of low blood pressure because they require different measures. The AI algorithms will initially be developed by collecting data from a large group of patients and then evaluating on a smaller group of other, representative patients Data from our standard monitoring in connection with surgery are supplemented with estimates of cardiac output based on the capnodynamic method. The research subjects will be put to sleep and ventilated via a modified intensive care ventilator (servo-i(R) Getinge). This ventilator has a software for research use that is CE marked. The software modifies the ventilator breathing pattern by to add slightly extended pauses after three out of nine breaths. This breathing pattern results in a mild variation in the level of exhaled carbon dioxide of about 1 kPa. The ventilator has been used in several large animal studies and clinical studies, two of which were conducted by our research group. Because an intensive care ventilator will to be used, the patients are anesthetized with a so-called total intravenous anesthesia (TIVA) instead of gas anesthesia. This is different from our current routine in this type of surgery. However, TIVA is used in several other types of surgery and we are well versed in the method. At several hospitals in the country and abroad, TIVA is used as first-line method even in large cancer surgery of the abdomen because there is data that suggests that patients receives minor relapses (metastasis) in the aftermath. During the course of care, the events that are recorded as occurs in order to then be added to the material for qualitative analyzes. Examples of such events are: administration of drugs and fluids, as well as probable cause of hypotension; bleeding body position changes, etc. To be able to compare the capno method's calculations of mixed venous saturation, blood samples will be drawn from the central venous catheter every hour during surgery. Each sample contains about 2 ml of blood. In a five-hour operation, this corresponds to about 1 cl of blood. Simultaneously with the vein samples, arterial blood gases will be taken with the same blood volumes. These samples are taken approximately: Each/every two hours during surgery in clinical routine but in the study protocol they will be taken every hour in sync with the samples from the central venous catheter. Also clinical outcome such as length of care and complications are saved so that the dose of hypotension can be linked to complications. What is described above are things that we measure and register in clinical routine and thus do not imply any further impact on the research subjects. Analysis of data is done afterwards by our researchers on our computers and servers. Through artificial intelligence (AI) and machine learning (ML) we will train AI algorithms with the goal of constructing Effective and accurate methods for predicting low blood pressure preventively. Because recorded data is used we can retrieve the outcome of the same data and compare our algorithms with what actually happened.