Nitric Oxide Levels Were Associated With Postoperative Delirium Following Cardiac Surgery

The study focused on adult patients undergoing cardiac surgery. Patients meeting the following criteria were excluded from the study: 1) age ≤ 18; 2) pregnant women; 3) individuals with preexisting preoperative delirium or a history of psychiatric disorders like schizophrenia and major depressive conditions; 4) those with an extensive history of psychotropic medications; 5) Use of nitrates after admission; and 6) patients with significant neurological disorders, such as severe traumatic brain injuries or recent strokes within the past 6 months. Blood samples were collected from patients at three different time points - pre-surgery after fasting, during surgery, and 24 hours post-surgery. These samples were stored in a central biobank at -80°C for future analysis. Details regarding the type and duration of anesthesia administered were also documented. Plasma nitric oxide (NO) levels were assessed using ethylenediaminetetraacetic acid whole blood samples (BTK074, Bio-swamp, Wuhan, China), with specific assay protocols accessible at https://www.bio-swamp.com/goodsDetail/BTK074.

Collection of research variables The research variables comprised nutritional indices, encompassing fasting blood glucose (GLU), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), triglycerides (TG), albumin (ALB), and lymphocyte count (LYC). Initial test results for these indices were obtained within 48 hours of each patient's hospital admission. In instances where data was missing, the corresponding values recorded in the patient's medical records within the past 6 months were utilized. Furthermore, additional clinical data included gender, age, comorbidities, and treatment protocols. Comorbidities assessed comprised hypertension, diabetes, chronic lung disease, and chronic renal failure. Chronic lung diseases encompassed chronic obstructive pulmonary disease, bronchial asthma, chronic bronchitis, and bronchiectasis.

Definition and setting of clinical outcomes The primary outcome, set as patients developing POD within 7 days post-surgery or before discharge, was assessed. In the derivation cohort, delirium diagnosis relied on documented POD in the medical records. If not explicitly stated, POD presence was determined based on the Confusion Assessment Method for Intensive Care Unit (CAM-ICU) results as recorded in nursing documentation. In the prospective cohort, trained physicians utilized the CAM-ICU protocol to assess patients' consciousness at 6-hour intervals for up to 7 days postoperatively, classifying a patient with positive CAM-ICU at least once during this period or before discharge as having POD.

Secondary outcomes included common postoperative adverse events such as prolonged mechanical ventilation (MV) (defined as continuous mechanical ventilation for ≥72 hours post-surgery), in-hospital all-cause mortality, and the occurrence of postoperative pulmonary complications (e.g., respiratory infection, respiratory failure, pleural effusion, pulmonary atelectasis, pneumothorax, bronchospasm, and aspiration pneumonia). Data collection and diagnostic processes were conducted independently by two researchers, with discrepancies resolved by a third researcher to ensure accuracy and consistency.

Machine Learning Processes Machine learning techniques were implemented using Python (version 3.13.0, Python Software Foundation). Data preprocessing commenced, starting with outlier processing using the 99% Winsorisation technique to address outliers' impact on the results. StandardScaler was then applied to normalize the dataset. Spearman's correlation coefficient was employed to evaluate the correlation among dataset features, visualized using a heatmap. The K-Means clustering algorithm, a pivotal unsupervised learning method, was utilized[15]. K-Means models were trained on derived cohorts, and the silhouette coefficient method determined the optimal cluster number for robust clustering outcomes. The validation set data underwent classification using the pre-trained K-Means model to predict cluster labels for each sample point.