No abstract

Background: Postoperative atrial fibrillation (PAF) is a frequent complication following cardiothoracic surgery and is associated with increased length of stay and increased morbidity. This study develops a novel multi-factorial computational model to predict PAF, as well as a simplified risk score for PAF based on this model. Methods: A multi-factorial computational model to predict PAF was trained on 592 patients in the MIMIC II database undergoing cardiothoracic surgery at the Beth Israel Deaconess Medical Center, using demographic, comorbidity, laboratory, and echocardiographic parameters. Parameters were chosen using LR stepwise-backward elimination applied to routinely collected covariates via logistic regression. A simplified risk score was created using the retained covariates based on the observed odds ratios (OR). Discrimination for both the model and risk score were calculated using the area under the receiver operating characteristic curve (AUC) via leave one out cross validation. Results: The multi-factorial computational model retained the following parameters: age in years (OR 1.06; p<0.001), serum chloride in mEq/L (OR 0.92; p=0.019), and left atrial diameter categorized from 0-3 (OR 1.84; p<0.001). The model achieved a significantly higher level of discrimination relative to any covariate alone (AUC 0.72; p<0.001 relative to all individual covariates). The simplified risk score is calculated as: +2 for age > 67, +1 for serum chloride < 105, +1 for each categorical increase in left atrial diameter (>40mm, >47.5mm, >52.5mm); and achieved comparable performance to the multi-factorial computational model (AUC 0.71; p<0.001) with a graded increase in risk (Figure). Conclusions: Multi-factorial computational modeling may provide a robust basis for discriminating between patients at high or low risk of PAF following cardiothoracic surgery. This model can be delivered as a simplified multi-factorial risk score for broad use by the bedside.