Predicting Postoperative Bleeding Complications with a Digital Real-Time Biomarker in Patients with Mechanical Circulatory Support
Purpose: Bleeding after cardiothoracic surgery affects the outcome of 2-6% of patients. The machine learning-based biomarker x-c-bleeding monitors the postoperative bleeding probability in real time and can predict bleeding requiring revision surgery, serving as an early warning system. But patients receiving mechanical circulatory support (MCS) exhibit distinct hemodynamic characteristics due to laminar flow, possibly affecting the validity of this digital biomarker.
Methods: We compared the validity of x-c-bleeding in 83 patients on MCS with 741 patients undergoing standard surgery (coronary artery bypass surgery N = 341; valve surgery N = 400). MCS patients with durable left ventricular assist device implantation (Heart Mate 3, Abbott, N = 21 (47%) or short-term MCS with concomitant thoracic surgery (N = 23 (53%); vaECMO (N = 6), Impella 5.5 (N = 11), ECMELLA (N = 6)) were classified as high-risk for relevant bleeding. MCS patients with only peripheral cannulation (vaECMO (N = 2), Impella 5.5 (N = 18), ECMELLA (N = 19)) were considered low-risk. The combined endpoint was defined as rethoracotomy for bleeding within 30 h after surgery and/or death due to hemorrhagic shock within 72 h. Differences between the area under the receiver operating characteristic curve (AUROC) for the last metric value were tested using DeLong’s method.
Results: The study population (male (N=606 [74%]), median age 64 years (IQR 57 - 72 years)) had a median predicted bleeding probability of 0.6% (IQR 0.1% - 4%). None of the low-risk stMCS patients met the endpoint; predicted bleeding probability was 0.2% (IQR 0.06% - 0.9%). Thirty-five patients (4%) reached the endpoint with predicted bleeding probability of 62% (IQR 18 - 79%): 10 (12%) high-risk MCS patients and 25 (3%) after standard surgery. The AUROC of x-c-bleeding for MCS patients was 0.908 ((95% CI 0.825 - 0.992); cut-off 0.2: specificity 0.82 (95% CI 0.73 - 0.9), sensitivity 0.8 (95% CI 0.5 - 1)), for high-risk MCS patients 0.847 (95% CI 0.71 - 0.99) and for standard surgical patients 0.836 (95% CI: 0.731 - 0.940), without a significant difference (p > 0.1).
Conclusion: The predictive capacity of x-c-bleeding for postoperative bleeding requiring revision surgery/death due to hemorrhagic shock was equally valid in patients with and without MCS, despite the absence of pulsatility and increased susceptibility to artifacts.
