Table 2.
Method and performance summary of the reviewed AI publications.
| Authors (et al.) | Year | Title | Data Source | Findings |
|---|---|---|---|---|
| Loghmanpour [57] | 2016 | A Bayesian model to predict RVF following LVAD Therapy | INTERMACS data Patients: 10,909 |
Systolic PAP, pre-albumin, LDH, and RV EF are the most predictive preoperative variables. AUC of acute, early, and late RHF predictions is between 0.83 and 0.90 with a sensitivity of 90% |
| Samura [58] | 2018 | Prediction of RVF after left LVAD implantation using ML for preoperative hemodynamics | Preoperative clinical and hemodynamic parameters Patients: 115 |
Prediction accuracy is 95%, AUC is 0.85 |
| Bellavia [59] | 2020 | Usefulness of regional RV and right atrial strain for the prediction of early and late RVF following a LVAD implant: a ML approach | Biomarkers, echocardiography, cath-lab measurements Patients: 74 |
Significant predictors: Michigan risk score, CVP, and systolic strain of RV free wall. ROC AUC is 0.95 |
| Shad [60] | 2021 | Predicting post-operative RVF using video-based deep learning | Preoperative echocardiography video Patients: 723 |
ML AUC is 0.729, CRITT AUC is 0.616, Penn AUC is 0.605 |
| Kilic [61] | 2021 | Using ML to improve risk prediction about durable LVAD implantation | INTERMACS data Patients: 16,120 |
48.8% and 36.9% in 90-day and 1-year mortality prediction improvements using ML compared with usual logistic regression data analysis |
| Kilic [62] | 2021 | ML approaches to analyzing adverse events following durable LVAD implantation | ENDURANCE trials Patients: 564 |
Bleeding, infection, and RHF are the most common postoperative adverse events. RHF has a strong transitive relationship with bleeding and infection |
| Nayak [63] | 2022 | ML algorithms that identify distinct phenotypes of RHF after LVAD implantation | IMACS data Patients: 2550 |
Four post-LVAD RHF phenotypes are identified Clinical outcomes are evaluated |
| Bahl [64] | 2023 | Explainable ML analysis of RHF after LVAD implantation | INTERMACS data Patients: 19,595 |
Five best predictors are identified Non-linear relationships are identified |
| Just [65] | 2023 | AI-based analysis of body composition that predicts the outcome for patients receiving long-term MCS | Preoperative CT scan Patients: 137 |
Adipose tissue is an indicator of postoperative major complications. |
Abbreviations: RVF: right ventricular failure; LVAD: left ventricular assist device; INTERMACS: Interagency Registry for Mechanically Assisted Circulatory Support; PAP: pulmonary artery pressure; LDH: lactate dehydrogenase; RV EF: right ventricular ejection fraction; AUC: area under the ROC curve; RHF: right heart failure; ML: machine learning; CVP: central venous pressure; ROC: receiver operating characteristic; IMACS: International Registry for Mechanically Assisted Circulatory Support; MCS; mechanical circulatory support; CT: computed tomography.