Table 1 ECMO-specific scores and their discrimination
From: Predictive models in extracorporeal membrane oxygenation (ECMO): a systematic review
| Â | Â | Â | Â | Internal validation | External | ||
|---|---|---|---|---|---|---|---|
Model name | Type of ECMO | Indication | Study design | Type of model | Variable selection | AUC (95% CI) | AUC (95% CI) |
6-h PREDICT VA-ECMO score [31] | VA | Cardiogenic shock | Retrospective, single centre | GLMNet | Elastic net | 0.82 (0.76–0.88) Cross-validation | 0.72 (0.65–0.78) |
12-h PREDICT VA-ECMO score [31] | VA | Cardiogenic shock | Retrospective, single centre | GLMNet | Elastic net | 0.84 (0.78–0.9) Cross-validation | 0.73 (0.65–0.82) |
AMI-ECMO score [32] | VA | Acute myocardial infarction | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable p-values; backward stepwise | 0.88 (0.82–0.94) LOO cross-validation | - |
CASUS score [33] | VA | Post-cardiotomy; circulatory failure | Retrospective, single centre | GLM (logistic) | - | 0.68 (0.57.0.77) | - |
Worku et al. score [34] | VA | Cardiogenic shock | Retrospective, single centre | GLM (logistic) | Based on univariable p-values | - | - |
New ECPR score [35] | VA | Cardiac arrest | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable and multiple regression p-value | 0.86 (0.77–0.94) Bootstrap | - |
12-month LOVE score [36]** | VA | ECMO for every condition | Retrospective, single centre | GLM (logistic) | Combined logistic and Cox | 0.75 (0.62–0.88) | - |
REMEMBER score [37] | VA | Cardiogenic shock following coronary artery bypass grafting | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable p-values; backward stepwise | 0.85 (0.79–0.91) Bootstrap | - |
SAVE score [38] | VA | Refractory cardiogenic shock | Retrospective based on ELSO registry | GLM (logistic) | Two steps: based on univariable and multiple regression p-value | 0.68 (0.64–0.71) Bootstrap | 0.90 (0.85–0.95) |
Modified SAVE score [39] | VA | ECMO for every condition | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable p-values; forward stepwise | 0.84 | Â |
ENCOURAGE score [40] | VA | AMI-related refractory cardiogenic shock patients | Retrospective, two centres | GLM (logistic) | Two steps: based on univariable p-values; backward stepwise | 0.84 (0.77–0.91) Bootstrap | - |
Siao et al. score [41] | VA | Cardiac arrest | Retrospective, single centre | GLM (logistic) | Based on univariable p-values | - Bootstrap | - |
Lee et al. score [42] | VA | Cardiac arrest | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable and multiple regression p-value | 0.86 (0.80–0.93) Bootstrap | - |
Simple cardiac score [43] | VA | Any form | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable and multiple regression p-value | 0.77 (-) | - |
ECMO-ACCEPTS score [44] | VA | Any form excluding respiratory leading cause | Retrospective, based on administrative data | Cox proportional hazards | Two steps: based on univariable and multiple PH | 0.64 (0.63–0.66) | - |
DNN score [45] | VA | Any form | Retrospective, single centre | Deep neural network | Learning vector quantization | 0.92 3 × 10-fold cross-validation | - |
SHOP score [46] | VA/VV | Any form | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable and multiple regression p-value | 0.87 (0.79–0.96)* | - |
RESP score [47] | VV | Respiratory failure | Retrospective based on ELSO registry | GLM (logistic) | Based on multiple regression p-value | 0.74 (0.72–0.76) Bootstrap | 0.92 (0.89–0.97) |
Modified RESP score [48] | VV | Respiratory failure | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable p-values; forward stepwise | 0.71 (0.63–0.78) | - |
PRESET score [23] | VV | Respiratory failure | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable p-values; backward stepwise | 0.85 (0.76–0.93) | 0.7 (0.56–0.83) |
Cheng et al. score [49] | VV | Respiratory failure | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable p-values; backward stepwise | 0.76 (0.67–0.85) | - |
Roch et al. score [50] | VV | ARDS | Retrospective, single centre | GLM (logistic) | Two steps: based on univariable and multiple regression p-value | 0.8 (0.71–0.89) | - |
ECMOnet score [51] | VV | ARDS due to H1N1 influenza | Prospective multi-centre | GEE | Two steps: based on univariable and multiple regression p-value | 0.86 (0.75–0.96) | 0.69 (0.56–0.83) |
Model 1 (Enger) score [52] | VV | Respiratory failure | Retrospective, single centre | GLM (logistic) | Clinical choice | 0.75 (0.69–0.8) Bootstrap | - |
Model 2 (Enger) score [52] | VV | Respiratory failure | Retrospective, single centre | GLM (logistic) | Clinical choice | 0.79 (0.73–0.84) Bootstrap | - |
PRESERVE score [53] | VV | ARDS | Retrospective multi-centre | GLM (logistic) | Two steps: based on univariable p-values; backward stepwise | 0.89 (0.83–0.94) | - |