Predictive Capability of Metabolic Panels for Postoperative Atrial Fibrillation in Cardiac Surgery Patients

• Published in: The Journal of surgical research Vol. 278; pp. 271 - 281
• Main Authors: Qian, Steve S., Crandell, Ian, Hanlon, Alexandra, Joseph, Mark, Poelzing, Steven
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01-10-2022
• Subjects: Atrial Fibrillation - diagnosis; Atrial Fibrillation - etiology; Cardiac surgery; Cardiac Surgical Procedures - adverse effects; Electrolyte imbalance; Humans; Logistic Models; Metabolic panels; Postoperative atrial fibrillation; Postoperative Complications - diagnosis; Postoperative Complications - etiology; Risk Factors; Metabolic panels; Cardiac surgery; Postoperative atrial fibrillation; Electrolyte imbalance
• ISSN: 0022-4804; 1095-8673
• DOI: 10.1016/j.jss.2022.04.061

Postoperative atrial fibrillation (POAF) occurs in up to 65% of cardiac surgery patients and is associated with an increased risk for stroke and mortality. Electrolyte disturbances in sodium (Na+), potassium (K+), total calcium (Ca2+), chloride (Cl−), and magnesium (Mg2+) are predisposing factors for POAF, but these imbalances are yet to be used to predict POAF. The purpose of this study is to determine whether the development of POAF can be predicted by blood plasma ionic composition.
Metabolic panels of patients with no prior history of atrial fibrillation who did (n = 763) and did not develop POAF (n = 2144) after cardiac surgery were obtained from the Carilion Clinic electronic medical record system. We initially evaluated serum Na+, K+, Ca2+, Cl−, and Mg2+ in the two groups using descriptive statistics via scatter and spaghetti plots and then with predictive modeling via logistic regression and random forest models.
Neither scatter nor spaghetti plots of electrolyte data revealed a significant difference between those who did and did not develop POAF. Two logistic regression models and two random forest models with POAF status as the outcome were generated using the first observation for each electrolyte and the coefficient of the linear regression, which was obtained from a linear fit of the scatter plot. The random forest model using the first observation had a sensitivity of only 12.2%, but all four models had specificities more than 97%.
Neither of the two logistic regression nor two random forest models were able to effectively predict the development of POAF from plasma ionic concentrations, but the random forest models effectively classified patients who would not develop POAF.