Incidence and pathophysiology of gastrointestinal bleeding during mechanical circulatory support: A retrospective analysis using machine learning algorithms

Background: End-organ hypoperfusion from cardiopulmonary shock may require mechanical circulatory support (MCS). However, patients receiving MCS risk the development of hemorrhagic complications, including gastrointestinal bleeding (GI). Examining potential risk factors for these complications improves clinical understanding. The purpose of this investigation was to study the risk for GI bleeding in MCS patients. Methods: Following IRB approval, patient characteristics, previously reported comorbidities, and the incidence of GI bleeding were reviewed from January 2017 to October 2023. Clinical variables underwent machine learning with autovalidation. Support vector machine modeling provided the best performance among the ensemble models tested. Results: In this study of 156 patients who underwent 284 MCS procedures, the incidence of GI bleeding was 6.0% CI 3.3–10.4%. Following machine learning, patients with insulin-dependent diabetes were associated with GI bleeding. The Receiver Operating Characteristic (ROC) curve demonstrated an area under the curve (AUC) of 0.85 with a misclassification rate of 7.5%. The relative risk of the need for major transfusion (>2 packed red blood cell units/episode) was 1.7 CI 1.1–2.5. The majority (87%), but not all, of these patients received unfractionated heparin therapy. Finally, hospital length of stay was increased in patients with GI bleeding. Conclusion: Insulin-dependent diabetes was associated with increased risk for GI bleeding during MCS, and these patients more often required major transfusions. Further evaluation of continuous anticoagulation therapy is warranted. Knowledge derived from this analytical study may guide the development of institutional protocols to improve care in this patient population.