A new machine learning algorithm to predict veno-arterial ECMO implantation after post-cardiotomy low cardiac output syndrome

Key Concepts in Machine Learning and Clinical Applications in the Cardiac Intensive Care Unit

PURPOSE OF REVIEW

Artificial Intelligence (AI) technology will significantly alter critical care cardiology, from our understanding of diseases to the way in which we communicate with patients and colleagues. We summarize the potential applications of AI in the cardiac intensive care unit (CICU) by reviewing current evidence, future developments and possible challenges.

RECENT FINDINGS

Machine Learning (ML) methods have been leveraged to improve interpretation and discover novel uses for diagnostic tests such as the ECG and echocardiograms. ML-based dynamic risk stratification and prognostication may help optimize triaging and CICU discharge procedures. Latent class analysis and K-means clustering may reveal underlying disease sub-phenotypes within heterogeneous conditions such as cardiogenic shock and decompensated heart failure. AI technology may help enhance routine clinical care, facilitate medical education and training, and unlock individualized therapies for patients in the CICU. However, robust regulation and improved clinician understanding of AI is essential to overcome important practical and ethical challenges.

Integrative Analyses of Single-Cell RNA-Sequencing and Bulk RNA-Sequencing Reveal Prognostic Markers of Peripheral Blood Mononuclear Cells in Patients with Cardiogenic Shock Receiving Venous-Arterial Extracorporeal Membrane Oxygenation Support Collected before Venous-Arterial Extracorporeal Membrane Oxygenation Establishment

The impact of changes in peripheral blood mononuclear cells (PBMCs) on the prognosis of patients with cardiogenic shock (CS) receiving venous-arterial extracorporeal membrane oxygenation (VA-ECMO) remains unclear. A single-cell RNA-sequencing (scRNA-seq) and a bulk RNA-sequencing (bulk RNA-seq) data sets of pre-ECMO PBMCs of CS patients were obtained from the gene expression omnibus database, which were analyzed using the "Seurat" and "limma" packages, respectively. The counts of different PMBC cell types, differential expression genes (DEGs), pathway enrichment analysis, cell-cell communication analysis, pseudotime analysis, and immune cell infiltration analysis were compared between VA-ECMO groups with different prognoses. The intersectional DEGs of the two data sets were screened. PBMCs were collected from VA-ECMO patients for experimental verification. For scRNA-seq analysis, ten kinds of PBMCs were identified, and B and NK/NKT cells which had significant differences in cell counts across groups were further divided into four subsets. The counts of B and NK/NKT cells with high expression levels of HSPA1B were higher in the poor-prognosis group, which was consistent with the bulk RNA-seq analysis. Pseudotime analysis also indicated that B-HSPA1B cells gradually increased in the poor-prognosis group. HSPA1B was found to be the intersectional upregulated DEG in both data sets, which was consistent with the experimental verification using clinical samples. The increased counts of B and NK/NKT cells as well as high expression levels of HSPA1B in these cell types or in the total pre-ECMO PBMCs were predictors of poor prognosis.

Machine learning-derived multivariable predictors of postcardiotomy cardiogenic shock in high-risk cardiac surgery patients

Objective

To develop a model for preoperatively predicting postcardiotomy cardiogenic shock (PCCS) in patients with poor left ventricular (LV) function undergoing cardiac surgery.

Methods

From the Society of Thoracic Surgeons Adult Cardiac Database, 11,493 patients with LV ejection fraction ≤35% underwent isolated on-pump surgery from 2018 through 2019, of whom 3428 experienced PCCS. In total, 68 preoperative clinical variables were considered in machine-learning algorithms trained and optimized using scikit-learn software.

Results

Compared with patients with ideal recovery, those that did were younger (65 vs 67 years), more likely female, Black, with low LV ejection fraction (26.5 vs 28.9%), previous myocardial infarction, chronic lung disease, diabetes, reoperation, or advanced heart failure. Among those with PCCS versus ideal recovery, operative mortality was 27% (925/3428) versus 0.1% (5/8065). PCCS occurred more often after coronary artery bypass grafting with concomitant mitral valve repair or after longer perfusion and clamp times. Reliable preoperative PCCS predictors were more advanced cardiac, liver, and renal failure; frailty; and greater white cell count. Out of sample test set receiver operating curve achieved an area under the curve of 0.74 with acceptable calibration Hosmer-Lemeshow statistic χ2 = 1.33, P = .25.

Conclusions

In patients with severe LV dysfunction undergoing cardiac surgery, risk of PCCS is elevated by preoperative failure of other organ systems and complexity of the planned operation that prolongs myocardial ischemia and cardiopulmonary bypass. This risk calculator could serve as an important tool to preoperatively identify patients in need of advanced levels of support.

Machine learning based predictive modeling of readmissions following extracorporeal membrane oxygenation hospitalizations

Background

Despite increasing utilization and survival benefit over the last decade, extracorporeal membrane oxygenation (ECMO) remains resource-intensive with significant complications and rehospitalization risk. We thus utilized machine learning (ML) to develop prediction models for 90-day nonelective readmission following ECMO.

Methods

All adult patients receiving ECMO who survived index hospitalization were tabulated from the 2016-2020 Nationwide Readmissions Database. Extreme Gradient Boosting (XGBoost) models were developed to identify features associated with readmission following ECMO. Area under the receiver operating characteristic (AUROC), mean Average Precision (mAP), and the Brier score were calculated to estimate model performance relative to logistic regression (LR). Shapley Additive Explanation summary (SHAP) plots evaluated the relative impact of each factor on the model. An additional sensitivity analysis solely included patient comorbidities and indication for ECMO as potential model covariates.

Results

Of ∼22,947 patients, 4495 (19.6 %) were readmitted nonelectively within 90 days. The XGBoost model exhibited superior discrimination (AUROC 0.64 vs 0.49), classification accuracy (mAP 0.30 vs 0.20) and calibration (Brier score 0.154 vs 0.165, all P < 0.001) in predicting readmission compared to LR. SHAP plots identified duration of index hospitalization, undergoing heart/lung transplantation, and Medicare insurance to be associated with increased odds of readmission. Upon sub-analysis, XGBoost demonstrated superior disclination compared to LR (AUROC 0.61 vs 0.60, P < 0.05). Chronic liver disease and frailty were linked with increased odds of nonelective readmission.

Conclusions

ML outperformed LR in predicting readmission following ECMO. Future work is needed to identify other factors linked with readmission and further optimize post-ECMO care among this cohort.

Overview and Clinical Applications of Artificial Intelligence and Machine Learning in Cardiac Anesthesiology

Artificial intelligence- (AI) and machine learning (ML)-based applications are becoming increasingly pervasive in the healthcare setting. This has in turn challenged clinicians, hospital administrators, and health policymakers to understand such technologies and develop frameworks for safe and sustained clinical implementation. Within cardiac anesthesiology, challenges and opportunities for AI/ML to support patient care are presented by the vast amounts of electronic health data, which are collected rapidly, interpreted, and acted upon within the periprocedural area. To address such challenges and opportunities, in this article, the authors review 3 recent applications relevant to cardiac anesthesiology, including depth of anesthesia monitoring, operating room resource optimization, and transthoracic/transesophageal echocardiography, as conceptual examples to explore strengths and limitations of AI/ML within healthcare, and characterize this evolving landscape. Through reviewing such applications, the authors introduce basic AI/ML concepts and methodologies, as well as practical considerations and ethical concerns for initiating and maintaining safe clinical implementation of AI/ML-based algorithms for cardiac anesthesia patient care.

Postcardiotomy Shock Syndrome: A Narrative Review of Perioperative Diagnosis and Management

Postcardiotomy shock (PCS) is generally described as the inability to separate from cardiopulmonary bypass due to ineffective cardiac output after cardiotomy, which is caused by a primary cardiac disorder, resulting in inadequate tissue perfusion. Postcardiotomy shock occurs in 0.5% to 1.5% of contemporary cardiac surgery cases, and is accompanied by an in-hospital mortality of approximately 67%. In the last 2 decades, the incidence of PCS has increased, likely due to the increased age and baseline morbidity of patients requiring cardiac surgery. In this narrative review, the authors discuss the epidemiology and pathophysiology of PCS, the rationale and evidence behind the initiation, continuation, escalation, and discontinuation of mechanical support devices in PCS, and the anesthetic implications.

Angiopoietin-2 is associated with capillary leak and predicts complications after cardiac surgery

BACKGROUND

Patients undergoing cardiac surgery are prone to numerous complications. Increased vascular permeability may be associated with morbidity and mortality due to hemodynamic instability, fluid overload, and edema formation. We hypothesized that markers of endothelial injury and inflammation are associated with capillary leak, ultimately increasing the risk of postoperative complications.

METHODS

In this prospective, observational, multidisciplinary cohort study at our tertiary academic medical center, we recruited 405 cardiac surgery patients. Patients were assessed daily using body impedance electrical analysis, ultrasound, sublingual intravital microscopy, and analysis of serum biomarkers. Multivariable models, as well as machine learning, were used to study the association of angiopoietin-2 with extracellular water as well as common complications after cardiac surgery.

RESULTS

The majority of patients underwent coronary artery bypass grafting, valvular, or aortic surgeries. Across the groups, extracellular water increased postoperatively (20 ± 6 preoperatively to 29 ± 7L on postoperative day 2; P < 0.001). Concomitantly, the levels of the biomarker angiopoietin-2 rose, showing a strong correlation based on the time points of measurements (r = 0.959, P = 0.041). Inflammatory (IL-6, IL-8, CRP) and endothelial biomarkers (VE-Cadherin, syndecan-1, ICAM-1) suggestive of capillary leak were increased. After controlling for common risk factors of edema formation, we found that an increase of 1 ng/mL in angiopoietin-2 was associated with a 0.24L increase in extracellular water (P < 0.001). Angiopoietin-2 showed increased odds for the development of acute kidney injury (OR 1.095 [95% CI 1.032, 1.169]; P = 0.004) and was furthermore associated with delayed extubation, longer time in the ICU, and a higher chance of prolonged dependence on vasoactive medication. Machine learning predicted postoperative complications when capillary leak was added to standard risk factors.

CONCLUSIONS

Capillary leak and subsequent edema formation are relevant problems after cardiac surgery. Levels of angiopoietin-2 in combination with extracellular water show promising potential to predict postoperative complications after cardiac surgery.

TRIAL REGISTRATION NUMBER

German Clinical Trials Registry (DRKS No. 00017057), Date of registration 05/04/2019, www.drks.de.