Table 3 Modeling strategies adopted in the selected studies
Author/year | Missing data | Model development | Model performance | |||||
|---|---|---|---|---|---|---|---|---|
Model type | Predictor selection method | Model format | Validation methods | Calibration | Discrimination | Classification | ||
Sico/2017 [23] | No imputation | Logistic regression | Backward with uniform P-value | / | External validation | / | D-C: 0.732 V-C: 0.731 | D-SN: 91.4%; SP: 43.8%; NPV: 76.2%, PPV: 72.1%; V-SN: 100%; SP: 12.5%; NPV: 100%; PPV: 79.6% |
Brown/2020 [25] | Separate category by default | Machine learning | Stepwise selection | / | / | / | C: 0.75 | / |
Boulos/2019 [29] | / | Logistic regression | / | / | Bootstrapping | / | C: 0.751 | SN: 95.9%; SP: 26.1%; PPV: 48.4%; NPV: 89.7% |
Katzan/2016 [27] | Multiple imputation | Logistic regression | Bootstrapping | Formula | Bootstrapping | / | STOP-BAG2 + C: 0.84 | SN: 94%; SP: 60% |
Bernardini/2021 [21] | / | Convolutional deep learning | / | / | / | / | / | / |
Zhang/2019 [22] | / | / | / | / | / | / | AUC: 0.835 | SN: 74.1%; SP: 76.9%; PPV: 87.5%; NPV: 57.7% |
Boulos/2016 [24] | / | Logistic regression | / | / | / | / | AUC: STOP-BAG: 0.677; 4 V: 0.688; BQ: 0.563; SOS: 0.506 | 4 V: SN: 59.4%; SP: 59.5%: PPV: 55.9%; NPV: 62.9% |
Petrie/2021 [26] | / | / | / | / | / | / | C-statistic SB, 0.572; ESS, 0.502; BQ, 0.640 | BQ: SN: 36%; SP: 62% ESS: SN: 68%; SP: 62% SB: SN: 81%; SP: 33% |
Šiarnik/2020 [31] | / | Logistic regression | Stepwise selection | / | / | / | AUC: 0.81 | SN: 82.9%; SP: 71.9%, |
Camilo/2014 [28] | / | Logistic regression | / | / | / | / | AUC: 0.813 | SN: 90%; NPV: 94.5%; SP: 55.6%, PPV: 27.1% |
Srijithesh/2011 [30] | / | / | / | / | / | BQ: SN: 68.2%, SP: 58.8%, PPV: 68.2%, NPV: 58.8% Combined BQ & ESS: SN: 50%, SP: 88.2%, PPV: 84.6%, NPV: 57.7% | ||