40 K. Gore et al.: J Extra Corpor Technol 2026, 58, 39 – 42
Table 1. Baseline characteristics and previously reported co-morbidities for gastrointestinal bleeding in 156 patients requiring mechanical circulatory support.
Terms |
Estimates |
Std error |
Chi-square |
p-values |
Intercept |
– 4.64 |
2.49 |
3.5 |
0.0625 |
Age |
0.03 |
0.03 |
1.4 |
0.2297 |
Sex [ female ] |
– 1.12 |
0.69 |
2.7 |
0.1030 |
BMI |
– 0.07 |
0.07 |
1.1 |
0.3045 |
Insulin-dependent diabetes |
3.51 |
1.54 |
5.2 |
0.0228 * |
Chronic renal failure |
1.05 |
0.72 |
2.2 |
0.1405 |
Chronic cardiovascular disease |
0.29 |
0.46 |
0.4 |
0.5281 |
Immunomodulation |
0.67 |
0.64 |
1.1 |
0.2948 |
Structural lung disease |
– 0.51 |
0.44 |
1.3 |
0.2545 |
ICD / Pacemaker |
0.23 |
0.52 |
0.2 |
0.6629 |
Atrial fibrillation |
0.52 |
0.52 |
1.0 |
0.3169 |
Previous cardiac surgery |
– 0.14 |
0.43 |
0.1 |
0.7408 |
Congestive heart failure |
– 0.29 |
0.49 |
0.4 |
0.5537 |
BMI: Body mass index( kg / m 2); ICD: Internal cardiac defibrillator; c-index statistic = 0.82; Misclassification rate = 6.0 %.
Results
Figure
1. Hospital length of stay in MCS patients with GI bleeding.
This complex relationship between MCS physiology and GI bleeding warrants further investigation. Therefore, the purpose of this analytical study was to examine the preprocedural risk factors for GI bleeding in patients receiving MCS.
Material and methods
Following IRB approval to allow electronic health records review without prior consent, data were entered into a study database. We conducted a retrospective single-center study of 156 patients who received 284 MCS devices( Appendix, Table A1) between January 2017 and October 2023 at Ochsner Health – Jefferson Highway Campus, New Orleans, Louisiana. Patients 18 years were included. No exclusion criteria were applied.
Statistics
Preprocedural comorbidities and patient characteristics( Table 1) underwent machine learning with autovalidation. Model performance was assessed using AUC and misclassification rates. Support Vector Machine modeling provided the best fit. JMP Pro 18.2 was used for all statistical analyses.
The incidence of GI bleeding was 6.0 %( CI 3.3 – 10.4 %). Insulin-dependent diabetes was associated with increased risk of GI bleeding( Table 1). Our institution applies insulin-guided protocols for glycemic management before and during MCS. The ROC curve for the Support Vector Machine model( Appendix, Figure A1) demonstrated an AUC of 0.85 and a misclassification rate of 7.5 %. The curve shows sensitivity versus 1-specificity. The steep initial rise indicates strong discriminative power, and the AUC demonstrates performance substantially above random( grey diagonal). The step-like pattern reflects limited data, highlighting the need for validation in larger cohorts. Patients with GI bleeding were more likely to require major transfusion(> 2 units pRBC / episode), with a relative risk of 1.7( CI 1.1 – 2.5). Most( 87 %) received unfractionated heparin therapy in accordance with institutional guidelines. GI bleeding was also associated with prolonged hospital length of stay( Figure 1).
Discussion
This investigation is, to our knowledge, the first report of increased GI bleeding among insulin-dependent diabetic patients receiving MCS across a range of device types. MCS confers substantial bleeding risk [ 6 ], and comorbidity profiles should guide individualized therapies. Further investigation of anticoagulation protocols, particularly continuous therapy, is warranted. GI bleeding was defined as hematochezia, hematemesis, melena, bloody NG output, transfusion requirement, or active bleeding observed at endoscopy. Patients with GI bleeding had nearly twice the relative risk of requiring major transfusion [> 2 units of packed red blood cells( pRBC)]. Transfusion may control bleeding but can worsen systemic inflammation, particularly in diabetic patients with baseline mucosal vulnerability [ 12, 13 ]. Prior studies, including Kapuria et al., have demonstrated higher rebleeding rates in diabetic patients on MCS [ 10 ].
In our cohort, GI bleeding was also associated with longer hospitalization, consistent with prior studies [ 6 ]. Together, these