Table 3 The characteristics of AI prediction models
First author and year | Algorithm/model | NO. of variables | Outcome(s) | Mode of validation | Missing data strategy | Best performing algorithm (Corresponding AUC) |
|---|---|---|---|---|---|---|
Crowe 2013 [27] | LR | 8 | 30-day mortality | Train/test | Record exclusion | LR (0.81) |
Zapata-Impata 2015 [28] | PSO + KNN | Combination A: 33 Combination A: 83 | RACHS classification | Train / Test | NP | PSO + KNN (NP) |
Moein 2015 [29] | RF | 1089 | Post-operative poor outcomes: Morbidity (need for ECMO, pLOS, etc.) and Mortality | Train /Test / Validation (SwB) | NP | RF with 400 trees + clinical features (0.743) |
Ruiz-Fernández 2015 [30] | MLP, SOM, RBF, DT | 87 | RACHS classification | 10-CV | Single imputation | MLP (0.999) |
Rogers 2017 [31] | LR | 11 | 30-day mortality | 5-CV + EV | Record exclusion | LR (0.86) |
Jalali 2018 [32] | SVM using 3 ranking methods: (1) mutual information (2) mutual information modified with reliability index (3) mutual information with reliability index and considering mutual information of a set | HLHS: 14 non-HLHS: 11 | Periventricular leukomalacia | K-fold cross validation | NP | SVM + 3rd ranking system (NP) |
Luis Ahumadal 2018 [33] | NN, RF | 26 | One-Year Transplant-Free Survival (Norwood Procedure) | 5-CV | Multiple imputation | NN (0.94) |
Samad 2018 [34] | LSVM | 24 (total): a: 6 b: 6 c: 5 d: 10 | a) Major vs No DVSF b) Major or Minor vs No DVSF c) Major vs Minor vs No DVSF d) Major vs Minor or No DVSF | 5-CV | NP | LSVM: a: 0.87 b: 0.82 c: 0.7 d: 0.77 |
Ruiz 2019 [35] | C-WIN + NB | 34 | Critical events: CPR, UETI, and ECMO in infants with SVP before second-stage surgery | 10-CV | Single imputation | C-WIN + NB (0.88) |
Cocomello 2020 [36] | LR | 11 | 30-day mortality | EV | Record exclusion | LR (1st cohort = 0.72; 2nd cohort = 0.88) |
Chang Junior 2020 [37] | MLP, RF, ET, SGB, ABC, BDT | MLP, BDT: 84; RF, SGB, ET, ABC: 42 | In-hospital /30-day mortality | 10-CV | NP | BDT (0. 926) |
Huang 2020 [38] | LR, NB, RF, LDA, SVM, KNN | 7 | Mean pulmonary arterial pressure > 15 mmHg | Train / Test | NP | RF (0.79) |
Bender 2021 [39] | SVM + Genetic Algorithm (the optimization technique) | 53 | Periventricular leukomalacia | Train / Test | NP | SVM (1) |
Bertsimas 2021 [40] | LR, OCT, RF, GB | 13 | In-hospital /30-day mortality pMVST pLOS | Train / Test | NP | Mortality: GB (0.874) pMVST: GB (0.856) pLOS: RF (0. 821) |
Faerber 2021 [41] | GB: two-stage with and without a MARS | without MARS: 56 with MARS: 21 | Postoperative cardiac complications | 10-CV | Multiple imputations | GB (0.71) |
Rusin 2021 [42] | LR | 7 | Cardiorespiratory deterioration events1 | Train / Test | NP | LR (0.958) |
Ng 2022 [43] | A deep learning based perioperative parameter classifier composed of CNN + RBF + fusion strategy | NA | - Length of ICU stay (LICUS) - Perioperative complications (PC) | Train /Test / Validation(SwB) | NP | LICUS: All components (0.73) PC: All components (0.72) |
Zeng 2021 [44] | XGBoost, LR | 45 | - Prediction of postoperative complications - Classification of postoperative complications | 5-CV | Multivariate imputation | Prediction: XGBoost (0.839) Classification: XGBoost (0.85) |
Thiriveedi 2021 [45] | XGBoost: STS model -Biomarker model -Clinical model | -STS model: NP -Biomarker model: 4 -Clinical model: NP | Readmission following 30Â days post operation | NP | NP | XGBoost: clinical model (0.997) |
Jalali 2021 [46] | LR, RF, DT, GB, DNN | 25 (1-year mortality) 49 (pLOS) | - 1-year mortality/need for cardiac transplant - Prolonged length of hospital stay (pLOS) | 5-CV | Multiple imputation | 1-year mortality: DNN (0.95) pLOS: DNN (0.94) |
Bertsimas 2022 [47] | OCT | 12 | - Mortality: In-hospital /30-day mortality - pMVST - pLOS | Train / Test | NP | OCT: Mortality: 0.872 OCT: pMVST: 0.814 OCT: pLOS: 0.813 |
Du 2022 [48] | XGBoost | 59 | In-hospital mortality | Train /Test / Validation(SwB) | NP | XGBoost (0.874) |
Ekhomu 2022 [49] | GB | NP | - Postoperative peak RA strain - Postoperative systolic RA strain rate - Postoperative early diastolic RA strain rate - Postoperative RV global longitudinal strain | 3-CV | Multiple imputations | GB (NP) |
Shi 2022 [50] | LR, SVM, MLP, XGBoost, AB | 15 | Malnutrition, defined as underweight: weight below −2 z-scores | K-fold CV + EV | Categorical: mode Continuous: multiple imputations | XGBoost (0.842) |
Pei 2022 [51] | Deep learning framework: 1. Segmentation and 3D modeling of LA and PV using V-net (CNN) 2. Computation of morphological features from LA and PV 3. Determination of Risk Factors 4. Risk Prediction Model by Morphological Features of LA and PV | 29 | Postoperative Pulmonary Vein Obstruction | 3-CV + EV | NP | CNN (0.870) |
Sunthankar 2023 [52] | LR, RF, XGBoost, GBDT, LightGBM | 180 | Interstage mortality between stage I and II surgery while at home | 5-CV | Single imputation | Light gradient boosting machine (0.642) |
Betts 2023 [53] | GBDT, ANN, LR | NP | 30-day mortality | 5-CV + EV | No missing data | Gradient boosting trees (0.87) |
Zürn 2023 [54] | LR | 5 | 30-day mortality | Leave-one-out CV + EV | Simple imputation | LR (0.9486) |
Jiwani 2023 [55] | CNN | NP | In-hospital mortality | K-fold CV | NP | CNN (NP) |
Kong 2023 [56] | LR, NB, XGBoost, SVM, LightGBM, MLP | 16 | Acute kidney injury | Train / Test / Validation (SwB) | NP | XGBoost (0.878) |
Sarris 2024 [57] | Decision trees | 12 | In-hospital mortality, pMVST, pLOS | Train / Test | NP | DT (Mortality: 0.866, pMVST: 0.851, pLOS: 0.818) |
Chang junior 2024 [58] | Catboost, RF, GB, NB, XGBoost, SVM, LightGBM. LR, DT, KNN, AB, LDA, ET, ridge, QDA | 93 | ICU length of stay | 10-CV | No missing data | Catboost (0.8559) |
Li 2024 [59] | LR, KNN | NP | Postoperative complications Mechanical ventilation duration | Train/ test | NP | LR + KNN (0.810) |
Smith 2024 [60] | NP | 45 to 195 (in different models) | Transplant-free survival | 5-CV | Missing forest | NP |
Tong 2024 [61] | LightGBM, LR, SVM, RF, CatBoost | 39 | LCOS, Pneumonia, Renal failure, Deep venous thrombosis | Train/ Test | Record exclusion | LCOS: LightGBM (0.893) Pneumonia: LR (0.929) Renal failure: LightGBM (0.963) DVT: LR (0.942) |