Journal: European Heart Journal. Digital Health

Article Title: Machine learning approaches for risk prediction after percutaneous coronary intervention: a systematic review and meta-analysis

doi: 10.1093/ehjdh/ztae074

Figure Lengend Snippet: Model characteristics

Article Snippet: Resnic et al. , LR, NN , Simplified prognostic risk scoring systems , 20 hidden nodes , Used forward and backward stepwise selection for model construction Excluded variables with high rates of missing data Conducted robustness checks on models without missing data adjustments Focused on maintaining a complete data set for key analysis variables , 67% training and 33% testing , ROC AUC: provided as C -statistic , NR , LR In-hospital mortality: AUC 0.85 In-hospital death, myocardial infarction, or same-day bypass surgery (MACE): AUC 0.78 NN In-hospital mortality: AUC 0.83 In-hospital death, myocardial infarction, or same-day bypass surgery (MACE): AUC 0.81 , LR Hosmer–Lemeshow In-hospital mortality: 13.060 NN Hosmer–Lemeshow In-hospital mortality: 7.170 Simplified prognostic risk scoring systems Hosmer–Lemeshow In-hospital mortality: 8.86 , Single random split (random split into training and test sets) , STATA 6.0 for multiple LR models and risk score models Goodman’s NetProv4 software for feedforward ANNs.

Techniques: Biomarker Discovery, Software, Transformation Assay, Selection