Development of an Artificial Intelligence Algorithm to Predict Hypotension Risk After Induction in Cesarean Sections With Spinal Anesthesia
Development of an Artificial Intelligence Algorithm to Predict Hypotension Risk After Induction in Cesarean Sections With Spinal Anesthesia
The cesarean section, medically necessary for both the mother and the baby in certain cases, is a life-saving operation.The most commonly used anesthesia method worldwide is spinal anesthesia. While spinal anesthesia has many advantages, it also has disadvantages. One of the most commonly encountered disadvantages is the development of hypotension due to the unopposed parasympathetic response after induction. Determining which patient will develop hypotension and which patient will not remains an important question for anesthesiologists before surgery. Identifying high-risk patients for hypotension before starting spinal anesthesia and even knowing the percentage of patients who will develop hypotension undoubtedly saves time in problem-solving. From this perspective, the idea for this study emerged: identifying parameters with the potential for use in prediction based on the literature, collecting data, then testing the relationship between them using machine learning methods, and developing an algorithm capable of predictive analysis.
At the end of the study, an artificial intelligence algorithm for predicting hypotension after induction will be developed, and its performance will be tested.
The main goals of the study:
i)Create a dataset including the clinical characteristics, demographic data, and blood test results of patients who develop and do not develop hypotension after spinal anesthesia.
ii) Develop an artificial intelligence algorithm using the dataset and determine the most accurate algorithm for predicting hypotension.
iii) To test the accuracy of the developed algorithm, create a test dataset, measure and optimize the algorithm's performance. Accuracy, sensitivity, specificity, and Receiver Operating Characteristic (ROC) curves will be used for performance measurement.
iv) Create a suitable interface (a surface for interaction with the software) to make the developed algorithm usable in clinical practice.
Rationale: Cesarean section, when indicated correctly, is a childbirth method that preserves the health of both the mother and the baby. The rates of births by cesarean section have been increasing worldwide for years. Between 2010 and 2018, 21.1% of globally tracked births were performed by cesarean section. According to the World Health Organization (WHO), it is expected that this rate will increase, reaching 29% by the year 2030.
The anesthesia for cesarean section is fundamentally influenced by both the physiological and pathological changes induced by pregnancy in the mother's body. Changes occurring during pregnancy and childbirth, and the resulting differences, can present challenges for anesthesiologists.
Spinal anesthesia induces iatrogenic sympathetic blockade, reducing systemic vascular resistance along with arterial and venous vasodilation, leading to hypotension. The incidence of hypotension after spinal anesthesia in pregnant women ranges from 7.4% to 74%. The frequency of hypotension is higher in pregnant women due to factors such as supine hypotension syndrome caused by fetal inferior vena cava compression and the development of collateral venous plexus in the epidural area, leading to the ascent of intrathecal local anesthetic in the cerebrospinal fluid. The deepening of hypotension and bradycardia in the patient can result in cardiac collapse, fetal hypoxia, and acidosis, posing an unpredictable risk to both maternal and fetal health. The role of anesthesiologists is to prevent or manage this risky condition effectively.
Recent advances in deep learning and artificial intelligence (AI) have found their place in the field of anesthesia. AI applications in anesthesia can be categorized into five main areas: 1) Monitoring the depth of anesthesia (e.g., techniques analyzing EEG data during anesthesia), 2) Control of anesthetic drug delivery based on depth of anesthesia, 3) Event prediction, 4) Ultrasound guidance, and 5) Pain management.
Among these applications, event prediction is particularly critical for anesthesiologists. Knowing about an event before it occurs contributes to its prevention or enables more accurate management. There have been 53 studies in the literature using AI for event prediction, including studies developing algorithms to predict hypotension during surgery and validating these algorithms. What sets this project apart from existing studies are:
These studies have not been specifically developed to predict spinal anesthesia-induced hypotension during cesarean sections.
These studies have mostly used wave analysis (non-invasive or invasive) for predicting hypotension. In this project, this approach will not be utilized. Preoperative assessments will be conducted, and patient characteristics will be recorded. Therefore, the resulting algorithm will not be a dynamic/real-time data input algorithm.
AI and its subset, machine learning algorithms, impact every aspect of our lives. In the medical field, it has become a method, especially in reducing human error in clinical decision-making.
In this study, demographic information, vital signs, specific blood parameters, and certain characteristics related to the administration of spinal anesthesia will be systematically recorded for pregnant women undergoing spinal anesthesia. Blood parameters include complete blood count, serum electrolytes, liver enzyme levels (ALT/AST), and Syndecan-1. All blood parameters are listed in the original study protocol. Syndecan-1 is a molecule shed into the serum with the degradation or damage of the vascular endothelial glycocalyx layer. Particularly, it rises during volume loading, sepsis, or inflammatory processes. Serum Syndecan-1 levels increase as the weeks of pregnancy progress, with the most significant increase occurring between the 20th and 30th weeks. If preeclampsia develops in the later stages of pregnancy, Syndecan-1 levels decrease.
In a study examining the preoperative and postoperative levels of Syndecan-1 in cesarean section patients who underwent spinal anesthesia, a significant difference was found. The study concluded that preoperative prophylactic fluid bolus administration affected endothelial glycocalyx degradation. Based on this study, Syndecan-1 has the theoretical potential to be a new marker for predicting hypotension. Although Syndecan-1 has been studied in pregnant women before, there is no study in the literature examining the relationship between preoperative Syndecan-1 levels and hypotension during spinal anesthesia in cesarean section patients. This study will both investigate the relationship between Syndecan-1 and hypotension and evaluate its place in the developed algorithm.
After recording the initial data for each patient, the development of hypotension will be observed and documented. The collected data will be analyzed to examine the relationships between the occurrence or non-occurrence of hypotension after induction. Additionally, an artificial intelligence algorithm will be developed using this data. The primary aim is to develop the algorithm. During the study, there will be no changes in the anesthesia management applied to the pregnant women.
In this context, the developed algorithm aims to early identify hypotension caused by spinal anesthesia (predicting the risk of hypotension before the operation starts) and enable early initiation of treatment.
Method:
Time of the Project: The ethical approval for the project has been granted by the Hacettepe University Non-Interventional Clinical Research Ethics Board. The project is planned to be conducted between November 2023 and April 2024.
Methodology and Data Collection Tools of the Project:
Data of patients meeting the inclusion criteria in the Operating Room will be obtained from routine test results sent after hospitalization for cesarean section, preoperative nurse observation forms, and records of vital signs monitored by anesthesia during surgery. After the patient is taken to the operating room, the information in the Data Collection Form will be filled out by the anesthesiologist. The duration of hypotension after spinal anesthesia begins immediately after anesthesia administration in pregnant women, reaching its lowest level within approximately 10-15 minutes. It is known to return to normal within 20-30 minutes. However, the duration of hypotension can vary depending on factors such as gestational age, anesthesia dose, overall health status of the patient, and other factors. Especially in pregnant women with high anesthesia doses or hypertension, the duration of hypotension can be longer . Therefore, the duration of hypotension needs to be evaluated individually for each patient. However, in this study, observation for the first 15 minutes is preferred. This preference is made because the algorithm predicts spinal-induced hypotension based on preoperative characteristics. With the extended observation period, the occurrence of childbirth, the increase in cardiac output, and the initiation of oxytocin infusion during the operation make hemodynamics a complex situation affected by multiple independent factors. Data collection will end after the first 15 minutes, and hypotension occurring after the 15th minute will not be recorded for the study; the observation will be terminated.
Collection of Project Data:
Before patients are taken to the operating room, an adequate amount of blood will be drawn from the routine pre-delivery blood samples taken from the patient for the blood parameters . Except for Syndecan-1, the other parameters in the blood parameters are routine tests routinely performed in the Hacettepe University Hospital Biochemistry laboratories.
When measuring syndecan-1, absorbance readings will be conducted using the SpectraMax-M2 (Molecular Devices, USA) device located in the Department of Biochemistry at Hacettepe University. Subsequently, serum syndecan-1 levels will be calculated using the GraphPad Prism program based on the standard graph. The procurement of the necessary ELISA kits for syndecan-1 measurement will be carried out through funding obtained from Hacettepe University's Scientific Research and Project Office.
Data Analysis:
Throughout the steps of algorithm development, the Python programming language will be used. The development process of the artificial intelligence algorithm will follow the Guidelines for Developing and Reporting Machine Learning Predictive Models in Biomedical Research: A Multidisciplinary View guideline. The steps are as follows:
Cross-Validation: The performance of the initial version of the model will be repeatedly subjected to cross-validation with subgroups of data that the model has never seen before. This will determine the model's performance by examining performance changes as the data varies. Depending on the data size, either 10-fold Cross-Validation or 5-fold Cross-Validation methods will be employed.
Internal Validation: The predictive performance of the algorithm trained with the training data will be assessed using internal validation data that it has never encountered.
In addition to the algorithm development process, if the variables being compared are normally distributed, statistical tests such as ANOVA; Student's t-test and Mann-Whitney U test will be used based on the number of compared groups and the normal distribution analysis of variables. Pearson Correlation Analysis and Spearman Correlation Analysis will be used to analyze the relationship between hypotension values and other continuous parameters based on the normal distribution analysis of variables. Moreover, the predictability of the system will be tested using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), odds ratio (OR), risk ratio (RR), receiver operating characteristic curve and area under the curve (AUROC), and Pearson correlation coefficient (r) tests. A statistical significance level of p<0.05 will be accepted. For statistical analysis, the Statistical Packages for the Social Sciences v26.0 (SPSS Inc., Chicago, IL) software will be used.
Inclusion Criteria:
Exclusion Criteria: