Pneumonia After Cardiovascular Surgery With Artificial Intelligence

Hospital-acquired pneumonia (nosocomial pneumonia) is one of the most common complications following cardiac and/or vascular surgery, with an incidence reaching up to 37% after certain types of procedures. This complication is associated with prolonged hospital stays, increased healthcare costs, and higher mortality rates both during hospitalization and within 5 years post-surgery. To determine the necessity of implementing additional personalized preventive measures against pneumonia, it is clinically valuable to assess the probability of its development in each individual patient. A medical risk calculator, which enables the calculation of an individual's risk of developing pneumonia, can serve as a supportive tool for clinical decision-making.

This is an observational study. Patient evaluation and treatment will be conducted in strict accordance with approved standards of medical care for the respective conditions. No experimental or unregistered (not approved for use) medical or diagnostic methods will be utilized in this study.

The study involves the systematic collection of data from the medical records of patients who undergo cardiac and/or vascular surgery during their current hospitalization. These data will be analyzed using artificial intelligence (AI) technologies, including machine learning methods such as gradient boosting, random forest, neural networks, Bayesian networks, and others.

This is a multicenter study involving three participating centers: Tomsk National Research Medical Center of the Russian Academy of Sciences, Tomsk State University, and the Federal State Budgetary Institution Research Institute for Complex Issues of Cardiovascular Diseases.

In phase 1, data will be collected from at least 400 patients undergoing cardiac surgery at a single center (Tomsk National Research Medical Center of the Russian Academy of Sciences). An anonymized database will be created, including the following parameters:

• Demographic and anthropometric characteristics (age, sex, height, weight, body mass index)
• Smoking status, including duration and intensity (pack-years)
• Comorbidities, including chronic obstructive pulmonary disease (COPD), diabetes mellitus, chronic kidney disease (CKD), prior myocardial infarction, and atrial fibrillation
• Preoperative left ventricular ejection fraction (LVEF)
• Preoperative pulmonary function test (PFT) parameters
• Preoperative length of hospital stay
• Intraoperative variables, including type of surgery, use and duration of cardiopulmonary bypass (CPB), volume of blood loss, blood component transfusion, duration of mechanical ventilation, and other relevant surgical details
• Perioperative laboratory data, including hemoglobin, red blood cell (RBC) count, white blood cell (WBC) count, platelet count, serum creatinine, and creatine phosphokinase (CPK) Subsequently, the occurrence or non-occurrence of hospital-acquired pneumonia will be documented within 30 days post-surgery. Additionally, other relevant clinical data regarding the course of the postoperative period will be collected.

Phase 2 will be conducted jointly by Tomsk National Research Medical Center of the Russian Academy of Sciences and Tomsk State University, focusing on data processing and analysis. This phase will include:

• Data integrity assessment and quality control
• Missing data imputation (handling of missing values)
• Development of predictive models for hospital-acquired pneumonia using multivariate binary logistic regression (SPSS version 26.0) and machine learning algorithms, including Bayesian networks, decision trees, random forest, gradient boosting, and recurrent neural networks (RNN).

Phase 3, also performed by Tomsk National Research Medical Center of the Russian Academy of Sciences and Tomsk State University, involves the evaluation and comparison of the developed predictive models. The models will be compared based on key performance metrics:

• Accuracy, sensitivity, and specificity
• Interpretability
• Robustness to overfitting and data incompleteness Based on this evaluation, the preferred predictive model for hospital-acquired pneumonia demonstrating optimal performance characteristics will be selected.

Phase 4 (conducted by Tomsk National Research Medical Center of the Russian Academy of Sciences and Tomsk State University) will focus on the development of a universal medical risk calculator. This tool will determine the individual probability of developing hospital-acquired pneumonia following any type of cardiac and/or vascular surgical intervention. The calculator will be constructed by selecting key predictors and evaluating the feature weights (coefficients) derived from the best-performing predictive model identified in Phase 3.

Phase 5 will involve the internal validation of the developed risk calculator. This phase will consist of the following steps:

• Prospective enrollment of an additional 200 patients into the observational study at the primary center (Tomsk National Research Medical Center of the Russian Academy of Sciences).
• Preoperative assessment of the individual risk (probability) of hospital-acquired pneumonia for each patient using the newly developed calculator.
• Surgical intervention followed by a 30-day postoperative observation period.
• Calibration and performance analysis by comparing the pre-test (predicted) probability of pneumonia with the actual observed postoperative outcomes.

Phase 6 will involve the external validation of the developed medical risk calculator across 1 to 3 cardiovascular surgery centers. Each participating center will conduct an evaluation identical to Phase 5, prospectively enrolling data from at least 100 patients. Initially, the Federal State Budgetary Institution Research Institute for Complex Issues of Cardiovascular Diseases will join the study, with the potential inclusion of additional medical centers at a later stage.

Phase 7 (conducted by Tomsk National Research Medical Center of the Russian Academy of Sciences and the Federal State Budgetary Institution Research Institute for Complex Issues of Cardiovascular Diseases) aims to evaluate the impact of hospital-acquired pneumonia on the development of clinical outcomes within 12 months post-cardiac surgery. These outcomes include:

• Primary composite cardiovascular and pulmonary outcome
• Secondary composite pulmonary outcome
• Secondary composite cardiovascular outcome

This phase will be conducted by gathering follow-up data from enrolled patients or their next of kin via in-person visits, telephone interviews, medical record reviews, or centralized electronic healthcare registries. Data will be collected on whether patients experienced any of the following events within 12 months after surgery:

• Pulmonary / Respiratory Events: Respiratory death, hospitalization for respiratory diseases, development of oxygen dependence, new-onset asthma, COPD, or interstitial lung disease, and initiation or intensification of bronchodilator or corticosteroid therapy.
• Cardiovascular Events: Cardiovascular death, acute myocardial infarction, unstable angina, myocardial revascularization, acute ischemic stroke, transient ischemic attack (TIA), acute heart failure, hospitalization for decompensated heart failure, new-onset atrial fibrillation or ventricular tachycardia, and initiation or intensification of antiarrhythmic therapy.

Phase 8 will be conducted by Tomsk National Research Medical Center of the Russian Academy of Sciences and Tomsk State University, focusing on the processing and analysis of the 12-month follow-up data. This final phase will include:

• Data integrity assessment and quality control.
• Missing data imputation for the long-term follow-up variables.
• Comparative analysis of the incidence of cardiovascular and pulmonary events within 12 months post-surgery between patients who developed hospital-acquired pneumonia during the early postoperative period and those who did not.
• Development of predictive models for the study's composite endpoints, utilizing both standard statistical methods and artificial intelligence (AI) / machine learning algorithms.