Prediction of Transfusions After Isolated Coronary Artery Bypass Grafting Surgical Procedures

Abstract

Background: Although blood transfusions are common and have been associated with adverse sequelae after cardiac surgical procedures, few contemporaneous models exist to support clinical decision making. This study developed a preoperative clinical decision support tool to predict perioperative red blood cell transfusions in the setting of isolated coronary artery bypass grafting.

Methods: We performed a multicenter, observational study of 20,377 patients undergoing isolated coronary artery bypass grafting among patients at 39 hospitals participating in the Michigan Society of Thoracic and Cardiovascular Surgeons Quality Collaborative's PERFusion measures and outcomes (PERForm) registry between 2011 and 2015. Candidates' preoperative risk factors were identified based on previous work and clinical input. The study population was randomly divided into a 70% development sample and a 30% validation sample. A generalized linear mixed-effect model was developed to predict perioperative red blood cell transfusion. The model's performance was assessed for calibration and discrimination. Sensitivity analysis was performed to assess the robustness of the model in different clinical subgroups.

Results: Transfusions occurred in 36.8% of patients. The final regression model included 16 preoperative variables. The correlation between the observed and expected transfusions was 1.0. The risk prediction model discriminated well (receiver operator characteristic [ROC]_development, 0.81; ROC_validation, 0.82) and had satisfactory calibration (correlation between observed and expected rates was r = 1.00). The model performance was confirmed across medical centers and clinical subgroups.

Conclusions: Our risk prediction model uses 16 readily obtainable preoperative variables. This model, which provides a patient-specific estimate of the need for transfusion, offers clinicians a guide for decision making and evaluating the effectiveness of blood management strategies.

Figure 1

Observed versus expected transfusion events by decile of predicted risk category. The correlation between the observed and expected events was 0.999.

Figure 2

Sensitivity analysis: center-level discriminatory ability. The mean, along with standard errors, are displayed for each medical center.